Uploaded by georgez1

978-3-030-97113-7

advertisement
Studies in Computational Intelligence 1033
Mariya Ouaissa · Inam Ullah Khan ·
Mariyam Ouaissa · Zakaria Boulouard ·
Syed Bilal Hussain Shah Editors
Computational
Intelligence
for Unmanned
Aerial Vehicles
Communication
Networks
Studies in Computational Intelligence
Volume 1033
Series Editor
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
The series “Studies in Computational Intelligence” (SCI) publishes new developments and advances in the various areas of computational intelligence—quickly and
with a high quality. The intent is to cover the theory, applications, and design methods
of computational intelligence, as embedded in the fields of engineering, computer
science, physics and life sciences, as well as the methodologies behind them. The
series contains monographs, lecture notes and edited volumes in computational
intelligence spanning the areas of neural networks, connectionist systems, genetic
algorithms, evolutionary computation, artificial intelligence, cellular automata, selforganizing systems, soft computing, fuzzy systems, and hybrid intelligent systems.
Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution, which enable both wide and rapid
dissemination of research output.
Indexed by SCOPUS, DBLP, WTI Frankfurt eG, zbMATH, SCImago.
All books published in the series are submitted for consideration in Web of Science.
More information about this series at https://link.springer.com/bookseries/7092
Mariya Ouaissa · Inam Ullah Khan ·
Mariyam Ouaissa · Zakaria Boulouard ·
Syed Bilal Hussain Shah
Editors
Computational Intelligence
for Unmanned Aerial
Vehicles Communication
Networks
Editors
Mariya Ouaissa
Moulay Ismail University
Meknes, Morocco
Mariyam Ouaissa
Moulay Ismail University
Meknes, Morocco
Syed Bilal Hussain Shah
Dalian University of Technology
Dalian, China
Inam Ullah Khan
Isra University
Islamabad, Pakistan
Zakaria Boulouard
Faculty of Sciences and Techniques
Mohammedia
Hassan II University
Casablanca, Morocco
ISSN 1860-949X
ISSN 1860-9503 (electronic)
Studies in Computational Intelligence
ISBN 978-3-030-97112-0
ISBN 978-3-030-97113-7 (eBook)
https://doi.org/10.1007/978-3-030-97113-7
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Unmanned aerial vehicles (UAVs), or aerial drones, have aroused an increasing
interest over the last few years. UAVs can be used in different fields, such as emergency communication networks, cartography, aerial photography, video transmission, military applications, modern agriculture, environment, parcel delivery, 3D
modeling of soils and buildings, and Wi-Fi broadcasting. The range of UAV use
cases and fields of application will be even wider thanks to the potential UAVs can
present.
Research or design in the field of UAVs is essentially multidisciplinary. Indeed,
the design of a UAV requires and calls upon knowledge of several disciplines, in
particular, electronics (especially on-board systems), automation, IT, mechanics,
aeronautics, robotics, and telecommunications.
Regarding communication areas, UAV-aided communication has been recognized
as an emerging and promising technique in industry for its superior on flexibility and
autonomy. To assist 5G communications, promising research scenarios can be as
follows: establishing temporal communication infrastructure during natural disasters,
offloading traffic for dense networks, and data collection/processing for supporting
Internet of Things (IoT) networks.
Recent progress in unmanned aerial vehicles and computational intelligence
constitutes a new chance for an autonomous operation and flight.
Nowadays, artificial intelligence and deep learning are driving the evolution of
UAVs and fuelling their autonomous future. Computer vision achieved a very important progress in image classification, segmentation, and object detection, which
makes it a very attractive research field when applied on unmanned aerial vehicles. As much as artificial intelligence can be an important and beneficial asset to
improve UAV performances, it can also be dangerous and a serious matter when the
UAVs learning is not managed correctly.
This book aims to provide a vision that can combine the best of both AI and
communication networks for designing the deployment trajectory to establish flexible
UAV communication networks.
This book is a collection of 16 original contributions that will discuss the major
challenges that can face deploying unmanned aerial vehicles in emergent networks. It
v
vi
Preface
will focus on possible applications of UAVs in a Smart City environment where they
can be supported by IoT, wireless sensor networks, as well as 5G, and beyond. This
book will discuss the possible problems and solutions and the network integration
of the UAVs, and compare the communication technologies to be used.
In Chapters Machine Learning and AI Approach to Improve UAV Communication and Networking, Implementation of Machine Learning Techniques in Unmanned
Aerial Vehicle Control and Its Various Applications, Machine Learning Techniques
for UAV Trajectory Optimization—A Survey and Metaheuristic Algorithms for Integrated Navigation Systems, the authors present a survey on different artificial intelligence and machine learning-based approaches to optimize UAV’s potential with
better communication, better control, and better trajectory management. In Chapters Security Threats in Flying Ad Hoc Network (FANET) and Secure Communication Routing in FANETs: A Survey, the authors present the concept of flying ad
hoc networks (FANETs) and different approaches to secure their communications;
while Chapter Impact of Routing Techniques and Mobility Models on Flying Ad Hoc
Networks focuses on the impact of these approaches on FANETs. Chapters Analysis
of Vulnerabilities in Cybersecurity in Unmanned Air Vehicles–Taxonomy of UAVs
GPS Spoofing and Jamming Attack Detection Methods try to analyze vulnerabilities in cybersecurity in UAV networks; while Chapter Investigation on Challenges
of Big Data Analytics in UAV Surveillance investigates the role of Big Data in UAV
surveillance.
The other chapters cover real-world success stories of the role of UAV in different
aspects of “Smart Cities”. In Chapters UAV-Based Photogrammetry and Seismic
Zonation Approach for Earthquakes Hazard Analysis of Pakistan–UAV-Based
Rescue System and Seismic Zonation for Hazard Analysis and Disaster Management, the authors provide concrete examples of the role of UAVs in the protection from natural disasters such as seism. Chapters Multi-sensor Fusion Methods
for Unmanned Aerial Vehicles to Detect Environment Using Deep Learning Techniques and General Parametric of Two Micro-Concentrator Photovoltaic Systems
for Drone Application try to cover the environmental aspect of UAVs either with
their energy management or their interaction with their surroundings.
Meknes, Morocco
Islamabad, Pakistan
Meknes, Morocco
Casablanca, Morocco
Dalian, China
Mariya Ouaissa
Inam Ullah Khan
Mariyam Ouaissa
Zakaria Boulouard
Syed Bilal Hussain Shah
Contents
Machine Learning and AI Approach to Improve UAV
Communication and Networking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bisma Baig and Abdul Qahar Shahzad
1
Implementation of Machine Learning Techniques in Unmanned
Aerial Vehicle Control and Its Various Applications . . . . . . . . . . . . . . . . . . .
E. Fantin Irudaya Raj
17
Machine Learning Techniques for UAV Trajectory
Optimization—A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sasikumar Rajendran, Karthik Kandha Samy,
Jeganathan Chinnathevar, and Deva Priya Sethuraj
35
Metaheuristic Algorithms for Integrated Navigation Systems . . . . . . . . . .
Ali Mohammadi, Farid Sheikholeslam, and Mehdi Emami
45
Security Threats in Flying Ad Hoc Network (FANET) . . . . . . . . . . . . . . . . .
Safia Lateef, Muhammad Rizwan, and Muhammad Abul Hassan
73
Secure Communication Routing in FANETs: A Survey . . . . . . . . . . . . . . . .
Shaheen Ahmad and Muhammad Abul Hassan
97
Impact of Routing Techniques and Mobility Models on Flying Ad
Hoc Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Muhammad Abul Hassan, Muhammad Imad, Tayyabah Hassan,
Farhat Ullah, and Shaheen Ahmad
Analysis of Vulnerabilities in Cybersecurity in Unmanned Air
Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Mohammad Ammar Mehdi, Syeda Zillay Nain Zukhraf,
and Hafsa Maryam
vii
viii
Contents
Silent Listening to Detect False Data Injection Attack
and Recognize the Attacker in Smart Car Platooning . . . . . . . . . . . . . . . . . 145
Sharmistha Majumder, Mrinal Kanti Deb Barma, Ashim Saha,
and A. B. Roy
Taxonomy of UAVs GPS Spoofing and Jamming Attack Detection
Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
A. Sabitha Banu and G. Padmavathi
Investigation on Challenges of Big Data Analytics in UAV
Surveillance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
N. Vanitha, G. Padmavathi, P. Nivedha, and K. Bhuvana
UAV-Based Photogrammetry and Seismic Zonation Approach
for Earthquakes Hazard Analysis of Pakistan . . . . . . . . . . . . . . . . . . . . . . . . 211
Abdul Qahar Shahzad and Mona Lisa
Optimizing UAV Path for Disaster Management in Smart Cities
Using Metaheuristic Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Zakria Qadir, Muhammad Hamza Zafar, Syed Kumayl Raza Moosavi,
Khoa N. Le, and Vivian W. Y. Tam
UAV-Based Rescue System and Seismic Zonation for Hazard
Analysis and Disaster Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Abdul Qahar Shahzad, Mona Lisa, Mumtaz Ali Khan, and Irum Khan
Multi-sensor Fusion Methods for Unmanned Aerial Vehicles
to Detect Environment Using Deep Learning Techniques . . . . . . . . . . . . . . 263
Pradeep Duraisamy, Venkatesh Babu Sakthi Narayanan,
Ramya Patturajan, and Kumararaja Veerasamy
General Parametric of Two Micro-Concentrator Photovoltaic
Systems for Drone Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Sarah El Himer, Mariyam Ouaissa, Mariya Ouaissa,
and Zakaria Boulouard
About the Editors
Dr. Mariya Ouaissa is a Researcher Associate and Practitioner with industry and
academic experience. She is a Ph.D. graduated in 2019 in Computer Science and
Networks, at the Laboratory of Modelisation of Mathematics and Computer Science
from ENSAM-Moulay Ismail University, Meknes, Morocco. She is a Networks and
Telecoms Engineer, graduated in 2013 from National School of Applied Sciences
Khouribga, Morocco. She is a Co-founder and IT Consultant at IT Support and
Consulting Center. She was working for School of Technology of Meknes Morocco
as a Visiting Professor from 2013 to 2021. She is member of the International
Association of Engineers and International Association of Online Engineering,
and since 2021, she is an “ACM Professional Member”. She is Expert Reviewer
with Academic Exchange Information Center (AEIC) and Brand Ambassador with
Bentham Science. She has served and continues to serve on technical program
and organizer committees of several conferences and events and has organized
many symposiums/workshops/conferences as a General Chair also as a reviewer
of numerous international journals. Dr. Ouaissa has made contributions in the fields
of information security and privacy, Internet of Things security, and wireless and
constrained networks security. Her main research topics are IoT, M2M, D2D, WSN,
cellular networks, and vehicular networks. She has published over 20 papers (book
chapters, international journals, and conferences/workshops), 5 edited books, and 5
special issue as guest editor.
Dr. Inam Ullah Khan was a Lecturer at different universities in Pakistan which
include Center for Emerging Sciences Engineering and Technology (CESET), Islamabad, Abdul Wali Khan University, Garden and Timergara Campus and University of
Swat. Recently, he is selected as a visiting researcher at King’s College London, UK.
He did his Ph.D. in Electronics Engineering from Department of Electronic Engineering, Isra University, Islamabad Campus, School of Engineering and Applied
Sciences (SEAS). He had completed his M.S. degree in Electronic Engineering at
Department of Electronic Engineering, Isra University, Islamabad Campus, School
of Engineering and Applied Sciences (SEAS). He had done undergraduate degree in
Bachelor of Computer Science from Abdul Wali Khan University Mardan, Pakistan.
ix
x
About the Editors
Apart from that, his Master’s thesis is published as a book on topic “Route Optimization with Ant Colony Optimization (ACO)” in Germany which is available on
Amazon. He is a research scholar; he has published some research papers at international level. More interestingly, he recently introduced a novel routing protocol
E-ANTHOCNET in the area of flying ad hoc networks. His research interest includes
network system security, intrusion detection, intrusion prevention, cryptography,
optimization techniques, WSN, IoT, UAVs, mobile ad hoc networks (MANETS),
flying ad hoc networks, and machine learning. He has served international conferences as technical program committee member which include EAI International
Conference on Future Intelligent Vehicular Technologies, Islamabad, Pakistan and
2nd International Conference on Future Networks and Distributed Systems, Amman,
Jordan, June 26–27, 2018, and now recently working on the same level at International Workshop on Computational Intelligence and Cybersecurity in Emergent
Networks (CICEN’21) that will be held in conjunction with the 12th International
Conference on Ambient Systems, Networks and Technologies (EUSPN 2021) which
is co-organized in November 1–4, 2021, Leuven, Belgium.
Dr. Mariyam Ouaissa is a Ph.D. Researcher Associate and Consultant Trainer in
Computer Science and Networks from Moulay Ismail University Meknes, Morocco.
She received her Ph.D. degree in 2019 from National Graduate School of Arts and
Crafts, Meknes, Morocco and her Engineering Degree in 2013 from the National
School of Applied Sciences, Khouribga, Morocco. She is a communication and
networking researcher and practitioner with industry and academic experience. Dr.
Ouaissa’s research is multidisciplinary that focuses on Internet of Things, M2M,
WSN, vehicular communications and cellular networks, security networks, congestion overload problem, and the resource allocation management and access control.
She is serving as a reviewer for international journals and conferences including as
IEEE Access, wireless communications, and mobile computing. Since 2020, she is
a member of “International Association of Engineers IAENG” and “International
Association of Online Engineering”, and since 2021, she is an “ACM Professional
Member”. She has published more than 20 research papers (this includes book chapters, peer-reviewed journal articles, and peer-reviewed conference manuscripts), 5
edited books, and 5 special issue as guest editor. She has served on Program Committees and Organizing Committees of several conferences and events and has organized
many symposiums/workshops/conferences as a General Chair.
Dr. Zakaria Boulouard is currently a Professor at Department of Computer Sciences
at the “Faculty of Sciences and Techniques Mohammedia, Hassan II University,
Casablanca, Morocco”. In 2018, he joined the “Advanced Smart Systems” Research
Team at the “Computer Sciences Laboratory of Mohammedia”. He received his Ph.D.
degree in 2018 from “Ibn Zohr University, Morocco” and his Engineering Degree
in 2013 from the “National School of Applied Sciences, Khouribga, Morocco”. His
research interests include artificial intelligence, big data visualization and analytics,
optimization and competitive intelligence. Since 2017, he is a member of “DraaTafilalet Foundation of Experts and Researchers”, and since 2020, he is an “ACM
About the Editors
xi
Professional Member”. He has served on Program Committees and Organizing
Committees of several conferences and events and has organized many symposiums/workshops/conferences as a General Chair. He has served and continues
to serve as a reviewer of numerous international conferences and journals. He
has published several research papers. This includes book chapters, peer-reviewed
journal articles, peer-reviewed conference manuscripts, edited books, and special
issue journals.
Dr. Syed Bilal Hussain Shah is currently Adjunct Professor at SKEMA Business
School Nanjing Audit University, China. He was Postdoctoral Researcher at the
School of Software, Dalian university of Technology, P.R. China. He has got Bachelor degree in Computer Sciences (2007) from University Department of Computer
Sciences University of Peshawar, Pakistan. He Joined Bahria University Islamabad,
Pakistan, for Masters in Telecommunication and Networking (2009). He completed
his Ph.D. from Dalian University of Technology, P.R. China. He has worked as
Lecturer at Department of Computer Sciences University of Peshawar (2010–2012)
Pakistan. He has authored/co-authored 50+ research papers in reputable journals
and conferences such as peer-to-peer networking and applications, future generation
computer systems IF, sustainable cities and societies, etc. Furthermore, published
papers in ACM, IEEE, and Springer conferences. Also presented his paper in conference, Cambridge, UK July 19, 2017. Main research interests include wireless sensor
network, IoT, throughput optimization in WSN, node localization, energy-efficient
routing in smart wireless sensor networks, distributed and centralized clustering in
WSN, IoT, blockchain, opportunistic networks, and Industry 4.0 technology.
Machine Learning and AI Approach
to Improve UAV Communication
and Networking
Bisma Baig and Abdul Qahar Shahzad
Abstract With the advent of unmanned aerial vehicles (UAVs) several sector of life
has been improved. Currently, numerous researches are carried out to enhance UAV
capabilities. UAVs are frequently utilized in several life-threatening operations such
as rescue, surveillance and transportation. Apart from this, drones-based experiments
are conducted in geology, wildlife, safety and ecological protection. Additionally,
5th generation approach which is consists of huge networks, high consistency and
transmission rates assist in UAVs. However, to attain such goals is business challenge for rapidly evolving Internet of Things (IoT), particularly in most dynamic and
mobile environments. Therefore, utilized in emergency where UAVs ensures rapid
recovery and tackling heavy traffic situations. These characteristics have attracted
the attention of organizations and academia. Moreover, machine learning (ML) and
artificial intelligence (AI) approaches are integrated into network where information
is used to solve problems. Thus, combination of ML with AI operates applications.
Furthermore, entire operation performance is enhanced. In this chapter, UAVs with
machine learning approaches are discussed. Study covers gaps in previous research
which influenced existing technique. In different context, machine learning (ML)
has recently become a subdomain of artificial intelligence.
Keywords IoT · UAV · AI · ML
1 Introduction
UAVs become very popular in recent years due to their basic characteristics which
include maneuverability, positioning, and capability to communicate with users
within line of sight (LOS). This aroused the interest of researchers. UAVs are basically
B. Baig
COMSATS University Islamabad, Abbottabad Campus, Islamabad 45550, Pakistan
A. Q. Shahzad (B)
Department of Earth Sciences, Quaid-I-Azam University, Islamabad 45320, Pakistan
e-mail: aqshahzad@geo.qau.edu.pk
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_1
1
2
B. Baig and A. Q. Shahzad
divided into two categories. One is fixed-wing UAVs and second is helicopter-based
UAVs. Both types of UAVs are suitable for a distinct type of application. As fixed
wing UAV is suitable for stationery-free missions, for example: military applications
like assault and reconnaissance. Nevertheless, the UAVs on the rotor have complicated aerodynamics. You can stay in certain locations as well, but you will not be able
to complete long-term missions. Furthermore, rotating-wing is ideal for providing
short-term reporting to operator. The engagement of different companies for aircraft
production carriers have reduced the market value, and the usage of UAV networks
doesn’t look like a dream or as an opinion for the future. Indeed, these are mostly
used for radio communications, agriculture, distribution, and reconnaissance. There
may be some restrictions on the use of UAVs. For example, visual LOS usage in
order to mitigates vulnerabilities in bad weather and dangerous loss of control. More
importantly, drone-limited batteries and low computing power are considered the
main limitations. In fact, UAVs on market find it difficult to take several hour flights
and return to base for batteries recharging. Moreover, restricted simulation power at
UAV prevents the execution of complex embedded algorithms that require high CPU
and GPU power. So, from scientific aspect complexity can affect its feasibility, so a
specific UAV problem is not always a good solution [1].
In addition, the large amount of data available today, multi-performance simulation (MPS) and the availability of better GPUs allowed AI to monitor brightness
of day. Consequently, machine learning used in many areas nowadays, even beyond
the expectation. Numerous sub-categories of AI are observed such as condensedlearning, deep-learning, and blended-learning for certain issues. For instance, artificial intelligence branch of DL utilize covers to mimic individual brain. Furthermore,
RL which is usually utilize in simulation, speech, and language recognition. This
part of AI introduced in 1979. Here, agents learn to do good deeds in order to receive
the greatest reward. The learning process can be accomplished by exploring and
exploring the various states available. Machine learning branch RL is considered to
evolve spontaneously as compared to other branches. Dissimilar to other, RL utilize
in field of robotics to learn how to plan routes and complete difficult jobs. Such
contribution of RL did not restrict the robotics field. Considerably, RL procedure
assists in main functions of making decision where target agents take into account
when intermingling with new prospective. However, FL is latest field in machine
learning primarily offered in 2016 by Google to sustenance distributed data. Machine
learning FL setup is consider for training highly centralized models on devices that
share distributed data without sending data to locally shared blocks. That is, utilize to
operate machine learning approaches in distributed information design. Therefore,
this technique considered as safe to do and using FL is a very hot topic for UAV-based
networks [2].
Machine Learning and AI Approach to Improve UAV Communication …
3
2 Literature Discussions
Neutrally, disparage some previous research, this portion mainly focus on the
machine learning as how it improve the general performance of UAV in issues
regarding wireless network. Firstly, machine learning tool is utilize to solve different
sort of complex issues and provide direction to tackle upcoming targeted problems.
Such techniques emphasis on the ML prospective in dealing with certain difficulties
where some literature, it is mixed with automatic learning [3].
Furthermore, the concept of solving machine learning is crystal clear as its finding
is more reliable and optimal when compared with other empirical approaches. Here
it advances some criticism: are ML techniques in dealing with problems better than
convectional approaches? Although previous successful experiments illustrates that
ML approach in solving issues is always superior than others, still there are many gaps
which needed to be fully addressed. While assessing the accuracy of ML technique
it is essential that data set is fully developed and having all the prospective which
needed to be solved. For instance, when working on the image to determine whether
it contain UAV or the number of nodes it contain with accurate positioning. This sort
of work is considered as visual detection of image through ML approach. Machine
learning investigates the image behavior through several methods to determine what
the image originally has? Here the proposed concept is utilized to find out the target
using the machine learning tool. In case the provided images have not numerous
drones. Definitely, CNN offers best way to find out through AI and ML approaches
[4].
Other shortcomings of ML can be found in several articles comparing methods
from a performance perspective. For example, when comparing the CNN architecture, when comparing two machine learning models, the calculation is to find out
why one model is better than another model and why the NN architecture is better
than the other. You may not see that there is no explanation. This point is typical
for ML black boxes. That is, it includes the parameters for adjusting and evaluating
the results and cannot be described elsewhere. As a result, it may not be possible to
predict which models will be used and which will not be promising for a particular
problem [8]. ML is another interesting option, however, especially for UAV problems. So I’m sure I can work on other ideas in the future. In fact, more complex
ML models can be tested with some UAV problems. For example, when predicting
path authorization, you can test several regression tools for this problem. I also know
that the UAV detection problem is solved by sound or image detection and this is
an automatic vision problem. On the other hand, with ML, if the NNs match (for
example, provide CNN type for images, RNN type for audio, radio), the last NN is
used to rank the output to estimate each input type. In addition, I noticed that I tend
to use supervised learning algorithms when troubleshooting UAV.
4
B. Baig and A. Q. Shahzad
3 UAVs Characteristics
As broadband request increases, universal reportage, access, local networks actively
sustenance existing terrestrial loopback-systems. A key component of NTN, the
Low Altitude Platform (LAP), aims to facilitate a variety of civilian, commercial
and governmental IoT missions and applications, from security operations and military services to amusement and broadcastings. Unmanned aerial vehicles, which
are the typical main type of CPU, are usually utilized for a short period of time
(numerous hours). This allows backbone of several link communications to be
quickly implemented in complex applications without public security personnel,
and rescue missions after natural disasters or unforeseen events, photo exploration.
Market research predicts UAV sales may approximately exceed $14 billion annually
by 2022. However, Federal Aviation Administration predicts number of UAVs may
reach to 2.4 million by 2022. The size and dynamics of the market are clearly the
motivating strength behind the development of UAVs. Recently, regulators, industry,
and academia have made a strong commitment to use drones as base stations,
mobile repeaters, or autonomous communications hubs to provide reliable, lowlatency communications in urban and suburban areas. The 3rd Generation Partnership Project (3GPP) [1] proposed aircraft utilization for long-term evolution and
IEEE 802.16 Relay Task Force introduced the concept of roaming relays. In 2013,
the Special Committee (SC-228) established a viable UAV framework established by
Radio Aeronautical Commission to formulate the technical characteristics of UAV
operations. In addition, in 2016, RTCA established the Unmanned Aerial Vehicle
Advisory Board to safely introduce unmanned aerial vehicles into the national
aerospace system. FAA and NASA launched combined initiative to assimilate UAVs
into US warfare systems [2]. From manufacturing perspective, main sellers such as
Microsoft, Google, Facebook, and YouTube tested 4G and 5G antenna platforms
for LTE uses. Depending on flight system, drones are divided in RPVs multi-rotor
drones (helicopter based drones), winged drones, front-wing/hybrid drones, robots,
and unmanned aerial vehicles. They range tiny toys to huge aircraft. In addition, UAVs
charge for communication equipment, cameras, radar, sensors, etc. from thousands
to milligrams in order to control the size and trip time. Due to exclusive features,
drone can achieve high coverage at high altitudes and high altitudes, and can provide
economical airflow in all areas using a large linear connection space (LoS) that moves
quickly in deployment and the movement is on demand [5].
In addition to using a small number of UAVs, UAV samples can work together to
perform complex tasks over very large areas, especially for monitoring and surveillance applications, but fly in special networks. (FANET) When most UAVs communicate ad hoc, connectivity and coverage can be achieved in situations of constraints
of the terrestrial network: remote sites, highly mobile and distributed sites. They can
be expanded effectively. However, while UAV interference must be effectively mitigated for successful UAVs operation built systems, UAV motion, supply supervision,
and governor are primarily based on UAV concentration. There is a problem with
the variety of types. The same is true for interoperability and positioning between
Machine Learning and AI Approach to Improve UAV Communication …
5
different wireless networks. In addition, not only is the UAV’s capacity limited in
terms of network load and onboard processing, but the need for engine power and
flying device is crucial real-world aspects limiting large-scale use of UAV.
In this regard, it is of great concern to extend the life of the drone, which is highly
dependent on flight performance and operation constraints. However, associated to
global wireless systems, UAV based network has unique and special structures which
include: topology, orbital paths and poorly attached nodes. Due to control constraints,
the energy-efficient model of onboard schemes requires route planning and battery
programming to extend range. In addition, mobility and Doppler forwarding can be
improved accordingly, and the quality of service (QoS) for information transmission
is distorted. In general, communications must be tailored to speed and QoS transfers
to attain preferred goals. Furthermore, current traditional communication methods
have fundamental limitations, especially in multipart situations where unpredicted
non-linear phenomena overcome. Consequently, mission-based, vibrant and acute
communications which lead to intricacy, ambiguity, and higher levels of unpredictability, the ML/AI has radically different decision-making abilities to find the
correct UAV location and trajectory. It is an important technology that it provides
[6].
4 Artificial Intelligence and Machine Learning
Artificial intelligence is considered to be scientific machines learning to execute
human-based responsibility. Artificial intelligence uses a variety of applications such
as speech recognition, robotic vehicles, machine-based interpretation, and communication. Furthermore, the techniques used to teach machines how to learn are a special
subset of artificial knowledge, latest structure called Machine Learning. ML offers
situations-based solution where numerous devices require simultaneous entrance to
system properties, for example during Internet of things data exchange. Intellectual
control across the network is required to meet the different needs of this new type
of service. The goal is to manage network resources as best as possible. Therefore,
machine learning procedures projected as effective method to solving conflicting
problems that arise from IoT bionetwork.
Generally, machine learning based on design background, the principal concept
of which used in correlations between previous datasets and a set of good deeds
to adapt to changes in the environment without human intervention. Obviously, the
advantage of wireless network machine learning frameworks is that network elements
can track, study, and forecast numerous communiqué factors which include traffic
patterns, wireless behavior, and tool locations.
DL is special machine learning class. DL uses multiple layers to create artificial
neural networks, allowing smart decisions to be made without humanoid interference.
The deep learning algorithms are used in restricted intervention needed but require
higher computing requirements. However, ML, DL and AI techniques are extensively
used in numerous wireless situations to improve many network parameters.
6
B. Baig and A. Q. Shahzad
4.1 Machine Learning Approaches
Machine learning falls into several categories, including controlled, semi-controlled,
and unendorsed learning with reinforcement [7].
4.1.1
Supervised Learning
Controlled based learning, algorithm uses a set of data. Both the required inputs and
outputs are available in this dataset. Therefore, this type of algorithm can only be
used when there is a large amount of tagged data.
4.1.2
Un-supervised Learning
Unconfirmed algorithm requires learning information, but lacks a labeled result. Such
sort of training performs pattern or collecting detection in the available data.
4.1.3
Semi-supervised Learning
The semi-supervised algorithm takes transitional method to type of data presented.
This type of training uses tagged and untouched data for training.
4.1.4
Reinforcement Learning
In RL, a series of actions using trial and error rules solves the problem. Thus,
basic concepts of training are very dissimilar from former idea of using historical
data. Instead, RL algorithm specializes in previous solutions on how to solve the
problem. The RL algorithm is used in a variety of situations in arena wireless network
optimization.
5 Unsupervised and Supervised ML for UAVs
Recent ML phrase associated to artificial intelligence. This subcategory of artificial intelligence allows computers perform accurately which is based on knowledge
expanded by studying few of the above cases. Machine learning experimentation is
conducted fruitful in previous period, thanks to amount of data accessible and simulation of today. Research is focused on relating machine learning to solve UAV-related
problems.
Machine Learning and AI Approach to Improve UAV Communication …
7
Fig. 1 Machine learning operational strategy types
ML areas are classified in dissimilar problem classes. For example, ML is
distributed into three categories such as supervised, unsupervised, and RL-based
issues as shown Fig. 1. Therefore, we will distinguish between knowledge areas to
avoid further misperception.
5.1 Supervised-Based Learning
Through this approach provided information is flagged. That is, it provides a true
baseline value for each data item so that the algorithm can learn to use those values
to determine new unlabeled items. For example, predict the price of a UAV based
on features. Here, algorithm is needed to offer along with dataset which include
the characteristics of individually drone with related. Datasets are generally divided
into training and testing sets. Supervised activities are often divided into graduate
activities or activities. Step-by-step activity remedies for the right to comment (such
as price forecasts).
However, the problem of classification provides cut-off values that indicate which
class an entry belongs to (e.g. classification of benign or malignant tumors). Below
is the most common machine learning statistic for controlled and unsubstantiated
learning. Also, it focuses on procedures utilize to solve UAV-related difficulties stated
in study.
8
5.1.1
B. Baig and A. Q. Shahzad
Related Division and Algorithms
Numerous algorithms are utilized for division and recession. For example, Support
Vector Machine performs dual approaches. You can also generate a decision tree to
resolve regressions or classifications depending on your use case.
5.1.2
Recession Algorithm
This approach utilizes algorithm which performs reversion. However, expecting
uninterrupted output of values is assigned.
5.1.3
Classifier Algorithm
Explain the basic concept of pure classifier Machine Learning. Some sources claim
that a naive Bayesian classifier with “some modifications” can be used for regression,
but since it was originally derived for Bayesian theorem classification.
5.1.4
Multilayer Perceptron (MLP)
RNA is mathematically formulated for machine learning to simulate human biological neural networks. ANN-based built-in on series of partly linked drones called
perceptron are gathered at dissimilar levels. However, detectors are accountable for
handling inbound and outbound delivery information.
5.1.5
Convolution Neural Networks (CNN)
Generally, CNN is broadly utilizes for the purpose of image detection. However,
method for understanding Natural Language Processing (NLP) and speech recognition is carried out. Where, the output and input layers are artificial neurons.
5.1.6
Recurrent Neural Networks (RNN)
If information is consistent, RNN is performed to correct the problem. For example,
you can browse text speeches, videos, or audio recordings. This approach is broadly
utilized for speech recognition, language interpretation, and language processing.
Machine Learning and AI Approach to Improve UAV Communication …
9
5.2 Unsupervised Learning Overview
Recent ML phrase associated to artificial intelligence. Below are the unsupervised
algorithms.
5.2.1
Clustering Algorithm
ML has some common aggregation algorithms. These are Gaussian Combustion
Module and Agglomeration. Algorithms based on density like DBSCAN, while
others act as strong associations like combustion. However, GMM is utilized to
modify association rules.
5.2.2
Dimension Reduction
Dimension reduction is classic machine learning method performed by altering
information from multidimensional 3D illustration to low dimension. Here, we
will discuss spectrum methods which include Auto encoder (AE), a type of neural
network used to explore and encode the representation of data. Surprisingly, the EC
architecture is very simple. Another common spectrum-based algorithm, principal
component analysis (PCA), is also referred to general procedure.
5.2.3
Generative Adversarial Networks
This approaches architecture utilized two basic system to create examples of information transmitted as real-time. It is typically used to create images, videos and
audio [8].
6 Solution for UAVs-Based Issues
Main issues in UAVs are related to its communications, coordination and interaction
with each other.
6.1 UAVs Coordination and Placement
The researchers monitor antenna stations to reduce the load on earth base stations
while minimizing drone power consumption. MAL assistance is considered a given
solution because the UAV does not have to keep moving forward, but is used
10
B. Baig and A. Q. Shahzad
temporarily to wait for the wireless network congestion. Expected congestion is due
GMMS. This could be a previously identified and unsupervised ML series model. For
example information utilizes the Gaussian scattering. Firstly, the averaging procedure distributes consumers in K-means groups, where maximum weight prediction
approach for group in order to achieve optimum factors of GMMM model. Furthermore, reduction of distribution is carried out through fixing UAV energy issue. Mathematical approach illustrate the finding of machine learning is more important than
the classical solution at the cost of reducing the mobility and energy required for
the links. Combining machine learning and optimization techniques is critical, but
when you use the K-mean algorithm to categorize users, you can manually select the
number of K groups and initiate a cluster center location. But the question is that,
“How to do it?” Several authors and researchers have explored the optimal placement of UAVs as base stations by constructing structured radio maps. Due to multifaceted behavior of arena with complication in radio map usage, the author proposes
grouping and joint regression problems using a maximum probability approach of
tracking design. Also, machine learning approach utilize channel forecast for radio
map reconstruction [9]. Use machine learning technology to predict loads with the
maximum weight expected. This machine learning technology is compared to the
maximum algorithm for optimizing expectations and the k-capacity algorithm for
performance. In addition, contract theory is used to ensure that load demand is met
by choosing the right UAV for each access point.
Researcher explored the creation of routes for UAV flight routes based on previous
flight records. The problem is that the mistakes made by pilots in the past will affect
future UAV flight routes. To this end, the authors applied an EU-based unsupervised
learning method to eliminate pilot errors and restore images generated from flight
recordings. The proposed method was compared with path generation using the K
resource algorithm and proved to be efficient. Efficiency amongst the base station
and the UAV enhanced through forecasting drone position in consideration of the
previous position. In fact, during the unloading of base stations on the ground, UAVs
can be exposed to wind turbulence, which causes delays and loss of capability. In
order to tackle such issues, researcher suggests (RNN) structure in which subsequent
parallel angle and height of drone respect to ground-station is projected through postangle. Such technique generates location-specific predictions in fast-moving UAVs.
The author constantly changes factors which include concealed number and level to
investigate predictions correctness. Mathematical approach illustrate correctness is
achieved in sixteen secret drones. Drone route plan are proposed in the published
literature [10].
6.2 Path Calculation
You may be wondering how ML can evaluate and model complex UAVs and established empirical models used to complicate UAV communication links. In this regard,
Machine Learning and AI Approach to Improve UAV Communication …
11
forecast of drone to drone route failure. Forecasts obtained through KNN (next neighborhood) and the Forest Random algorithm is compared with the statistical analysis.
Expected route failure is based on many factors which include length, height, and
altitude. Comparing results obtained with the ray tracing software, it can be seen that
ML did a good job with this planned activity. Because millimeter-wave bandwidth
is used in next-generation mobile systems to increase bandwidth, they used neural
networks to predict ATS channel conditions between two basic stations. Level Base
Station (ii) Roof antenna base station. The first neural network classifies the connection type (LOS/NLOS/off) sends this information to the second neural network to
generate different channel parameters.
Some authors used a GAN to characterize the air-ground link for millimeter-wave
communication in a wireless UAV network. The free distributed architecture aims
to provide training for distributed UAV networks. The training process is based on
a distributed data set to measure the channel. The machine-supervised study was
used in another study to predict the quality of the relationship between UAVs and
ground glands. For example, ANN is used to predict road losses for. ANN is used for
predicting UAV signal strength and speculates propagation in the channel. The ANN
scale is designed to analyze the effects of several common factors on signals, such as
contrast, reflection, and diffusion. The right side includes restrictions like distance
to the UAV, altitude, time, lost route, and more. This intriguing task could disrupt the
extensive data processing time at ANN. Meanwhile, signal power amongst drones
with base node is formulated through this approach. Author examines built-up setting
where signal power information is utilized in order to provide ANN flow. While
consuming information, factor is precisely predicted. Additionally, this approach
is utilized with machine learning strategy to forecast strength of the signal receive
through the aircraft from the mobile base station [11–13].
In short, an unsupervised study has been used to simulate 3D channels amongst
drone with cellphone operators. This problem uses Gaussian methodology to classify
LOS and NLOS references. This work uses means procedure to categorize LOS and
NLOS references.
6.3 Virtual Reality in Drones
This approach explains the real environments for various purposes which include
amusement and learnings. Machinery commenced to attract considerable attention
in recent times and recently observed an interesting line of research linking to merge
the virtual reality with drone system. Artificial intelligence utilized solves numerous
problems of implementing virtual reality. This approach acquires optimal transmission which includes small invisibility and great information proportions. Whereas,
some cases, researchers recommend a VRUAV evaluate efficiency of numerous
DL solutions. Some are investigating whether UAV-IoT networks could be used
to immerse themselves in virtual reality remotely. Unmanned aircraft are placed
12
B. Baig and A. Q. Shahzad
Fig. 2 Machine learning operational strategy
in areas of particular interest, assuming different perspectives and sending them to
specific aggregation points.
Accumulation is carried out to include virtual reality at operator geo-positioning.
Main objective of this VR is to enhance accuracy in specific circumstances. In this
task, we used the RL method to find the best UAV location for maximum immersion
accuracy. The DL method is transmitted and submitted, and RL approach associated
to the latest existing mathematical based method. Furthermore, obtained results were
compared where the former statistical approach show high efficiency [14].
6.4 Abnormalities in Drone Monitoring
UAVs often suffer from the sensitivity to any anomalies that might occur when a drone
is in operation. To avoid this situation, offered information through UAV sensor to
monitor flight safety level. However, irregularity finding is considered as common
procedure where involves identifying irregular information systems in the operating
system. In Fig. 2 some researchers devises an uncontrolled algorithm for detecting
adverse events that occur in drone testing [8].
6.5 UAVs Detection
As UAVs are used by both civilians and the military, authorities control certain applications in which UAVs can be used for espionage or as deadly weapons. Therefore,
UAV detection and monitoring is essential to prevent these threats. Many research
services in this field develop different methods, sharing them in image and audio
based solutions. One way to solve UAV detection problems is to use visual representation such as object displayed-images. Machine learning architecture of various
Machine Learning and AI Approach to Improve UAV Communication …
13
depths was compared with the visual detection task of a UAV through specific image
capturing tool [15]. Additionally, some articles explore some of the machine vision
techniques used to detect UAVs.
Finally, emphasize on creating cross scheme which utilize images, audio signals,
and radio together would be a very good idea in the future.
6.5.1
Drone Photogrammetry
Visual-based surveillance considered being outside this specific field, but there are
several studies in the literature related to UAV imaging. For example, detection of
mandatory (emergency) safely landing on sites. Discovery turns into cataloguing
issue in which dual well-known classifiers such as SVM and GMM are tested. These
turns into 3D view to secure/insecure lattice record. Furthermore, filtration is carried
out in order to eliminate the hazard of ground base station. Specifically, main objective
of this type problem was not measured in such interrogation that it can only be
considered visual-based surveillance but only image is not capture from specific
elevation, it could be some sort of task to apply to. In other words, techniques which
include feature-based extraction, CNN, and boundary indicators are used to capture
images from UAVs [16].
7 Interpretation and Future Practice
The use of machine learning technology in UAVs systems are problematic due to
restricted on-board computing capabilities. In fact, most off -the -shelf UAVs do
not require complex processors to execute complex machine learning procedures.
However, proposed UAV design with authoritative processor, simulator, and GPU,
you still need to consider their complete cost with weight. Consequently, UAV power
limitations continue to cause similar problems. By using the cloud to train models and
drawing conclusions at the UAV level, we can find a solution to the problem. Although
this specific elucidation enhances communiqué costs, and the UAV must alternately
communicate with the cloud, returning to the problem of energy constraints. So
using ML on board is another good solution, but this time set the ML algorithm to
UAV power limit. This approach takes us to a new area, commonly known as device
learning, especially for disabled devices.
In addition to the limitations of the UAV hardware and software cited overhead
where actual utilization of machine learning in Adhoc system poses numerous serious
obstacles to the current principals. Whereas, this study focused on the use of partial
or stand-alone UAVs, in general the existing regulatory requirements do not allow
for such real operations. However, this is crucial to note that there are areas which
need improvement. Dissimilar to FAA, latest regulation of the European Aviation
Safety Agency (EASA), published in December 2020, operates autonomous UAVs,
including classification according to level of application threat. Therefore, it ensures
14
B. Baig and A. Q. Shahzad
novel prospects for advancement in Adhoc network system elucidations based on
machine learning and artificial intelligence in general term [17].
Finally, this is significant to synchronize and integrate regulations for specific
application of UAVs around the world to stimulate future research [18].
8 Conclusion
To conclude, a machine learning framework whether it is supervised or unsupervised has tackles effectively several obstacles by providing intelligent-solutions to
a multitude of UAV-related challenges. Therefore, in the future, we believe that we
can continue to explore more supervised and unsupervised learning methods.
AI is one of the most popular parts where it informs cars and enables them to
perform tasks better than humans. Combining the benefits of using artificial intelligence in a UAV network is considered an interesting and inspiring idea. Although
traditional methods have been successful in solving many problems in this area, it
is still interesting to investigate whether ML and RL can provide more powerful
and accurate solutions. The transition from traditional learning to smart learning
may require sacrificing readability and management, but AI solutions, especially
given the unprecedented success of machine learning and smart learning in decisionmaking tasks. It is worth choosing. However, while we believe that smart solutions
do not always outperform traditional solutions, traditional types of approaches can
sometimes provide simple and effective solutions. Undoubtedly, this ambiguity is
one of the reasons for investigating the use of AI to solve some special problems in
UAV networks. Originally, UAVs were designed for full manual control of humans,
but with the recent advent of artificial intelligence, the market has offered intelligent
UAVs. In this context, artificial intelligence can use the information gathered by the
drone’s sensor to perform a multitude of tasks. Artificial intelligence can play an
important role in UAV resource management to maximize energy efficiency. Orbit
and UAV implementation projects are also subject to artificial intelligence improvements due to the ability of UAVs to avoid obstacles and automatically calculate their
orbit. For example, “Follow Me” drones have been very successful in the market
these days. This type of drone provides excellent video capture, owner surveillance,
and capture, and features powerful and intelligent algorithms to avoid obstacles. In
addition, this context allows you to extend a wide range of applications such as
surveillance, motion control, and landing detection. UAV visualization can also be
improved by using existing modern computerized visions for UAV imaging.
In a nut shell, the performance of a UAV-based network can be significantly
improved by integrating information algorithms to automate complex tasks and
increase the level of information in the system.
Machine Learning and AI Approach to Improve UAV Communication …
15
References
1. 3GPP TR 36.777 V1.1.0: study on enhanced LTE support for aerial vehicles. Tech. Specification
Group Radio Access Network (2018)
2. National Aeronautics and Space Administration (NASA), Unmanned Aircraft Systems Integration in the National Airspace system Project, (2016)
3. B. Li, Z. Fei, Y. Zhang, UAV communications for 5G and beyond: recent advances and future
trends. IEEE Internet Things J. (2019)
4. M. Mozaffari, W. Saad, M. Bennis, Y. Nam, M. Debbah, A tutorial on UAVs for wireless
networks: applications, challenges, and open problems. IEEE Commun. Surv. Tutor. (2019)
5. A. Fotouhi, H. Qiang, M. Ding, M. Hassan, L.G. Giordano, A. Garcia-Rodriguez, J. Yuan,
Survey on UAV cellular communications: Practical aspects, standardization advancements,
regulation, and security challenges. IEEE Commun. Surv. Tutor. (2019)
6. P. S. Bithas, mosthenes Vouyioukas, nasios G. Kanatas, A survey on machine-learning
techniques for UAV-based communications. MDPI (2019)
7. E. Alpaydin, Introduction to Machine Learning, (MIT Press, Cambridge, MA, USA, 2014)
8. M.-A. Lahmeri, M.A. Kishk, M.-S. Alouini, Artificial intelligence for UAV-enabled wireless
networks: a survey. IEEE Open J. Commun. Soc. 2, 1015–1040 (2021)
9. Q. Zhang, W. Saad, M. Bennis, X. Lu, M. Debbah, W. Zuo, Predictive deployment of UAV base
stations in wireless networks: machine learning meets contract theory. IEEE Trans. Wireless
Commun. 20(1), 637–652 (2021)
10. J. Kwak, Y. Sung, Autoencoder-based candidate waypoint generation method for autonomous
flight of multi-unmanned aerial vehicles. Adv. Mech. Eng. 11(6), 1–16 (2019)
11. S. Alsamhi, O. Ma, S. Ansari, Predictive estimation of the optimal signal strength from
unmanned aerial vehicle over internet of things using ANN
12. P. Ladosz, H. Oh, G. Zheng, W.-H. Chen, A hybrid approach of learning and model-based
channel prediction for communication relay UAVs in dynamic urban environments. IEEE
Robot. Autom. Lett. 4(3), 2370–2377 (2019)
13. S.K. Goudos, G. Athanasiadou, Application of an ensemble method to UAV power modeling
for cellular communications. IEEE Antennas Wireless Propag. Lett. 18(11), 2340–2344 (2019)
14. M. Chen, W. Saad, C. Yin, Deep learning for 360 content transmission in UAV-enabled virtual
reality, in Proceeding of IEEE International Conference on Communications (ICC), (2019),
pp. 1–6
15. J. Park, D.H. Kim, Y.S. Shin and S.-H. Lee, A comparison of convolutional object detectors for
real-time drone tracking using a PTZ camera, in Proceeding of IEEE International Conference
on Control, Automation and Systems (ICCAS), (2017), pp. 696–699
16. K. Mukadam, A. Sinh, R. Karani, Detection of landing areas for unmanned aerial vehicles,
in Proceeding of IEEE International Conference on Computing Communication Control and
automation (ICCUBEA), (2020), pp. 1–5
17. S. Dhar, J. Guo, J. Liu, S. Tripathi, U. Kurup, M. Shah, On-device machine learning: an
algorithms and learning theory perspective
18. V. Sharma, R. Kumar, K. Srinivasan, D.N.K. Jayakody, Coagulation attacks over networked
uavs: concept, challenges, and research aspects. Int. J. Eng. Technol. 7, 183–187 (2018)
Implementation of Machine Learning
Techniques in Unmanned Aerial Vehicle
Control and Its Various Applications
E. Fantin Irudaya Raj
Abstract An unmanned aerial vehicle (UAV), sometimes known as a drone, is an
aircraft or airborne system that is controlled remotely by an onboard computer or a
human operator. The ground control station, aircraft components, and various types
of sensors make up the UAV system. UAVs are categorized depending on their
endurance, weight and altitude range. They can be used for multiple commercial
and military applications. UAV intelligence and performance entirely depend on
their ability to sense and comprehend new and unfamiliar environments and conditions. Numerous Machine Learning (ML) algorithms have recently been developed
and implemented in the UAV system for this purpose. The integration of machine
learning and unmanned aerial vehicles has resulted in outputs that are both fast and
reliable. It will also lessen the number of real-time obstacles that UAVs confront while
simultaneously boosting their capabilities. Additionally, it will pave the way for the
application of UAVs in a number of different fields. The current chapter discusses
in detail machine learning approaches and their integration with unmanned aerial
vehicles. Additionally, it discusses the application of UAVs in various domains and
their effectiveness.
Keywords Unmanned aerial vehicle · Machine learning · Drone · Classification of
UAV · Recent trends and values · UAV applications
1 Introduction
Unmanned Aerial Vehicles (UAVs), or drones as they are commonly called, have
only been in use for around 60 years. Many countries’ air defences now include
unmanned aerial vehicles (UAVs). Since the US Air Force deployed unmanned
drones in the 1940s, unmanned aerial vehicles have made great strides. Those drones
were designed for observation and espionage; however, their operational systems
contained defects that rendered them useless [1]. After the continuous research and
E. Fantin Irudaya Raj (B)
Department of Electrical and Electronics Engineering, Dr Sivanthi Aditanar College of
Engineering, Tiruchendur, Tamilnadu, India
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_2
17
18
E. Fantin Irudaya Raj
revolution, UAVs have evolved into the very sophisticated devices they are today. In
the present world, UAVs can be used in various fields it varies from agriculture to
military applications [2].
The innovation in the UAV is started at the beginning of the 1900s. In the year, 1922
the first launch of UAV, namely RAE 1921 Target is getting launched from the British
aircraft carrier HMS Argus. The first successful flight of a radio-controlled UAV was
in the year 1924, which flew around 39 min. Starting from this, many countries all
around the world have started to show a high level of interest in unmanned aerial
vehicles. They started to invest huge amounts of money in research and development
to develop modernized and more reliable UAVs. In 1998, the first trans-Atlantic
crossing of UAV was performed. Trans-Pacific crossing of a UAV was performed in
the year 2001. These are a few important developments in the evolution of UAVs [3].
Figure 1 depicts the recent types of simple UAVs used in recent times.
The recent developments in the UAV make it viable for various applications.
Because UAVs can be controlled remotely and flown at varying distances and heights,
they are great candidates for tackling some of the world’s most demanding jobs.
Drones can be spotted aiding in the looking for survivors following a cyclone,
supporting police departments and the army with an eye in the sky amid terrorist
occurrences, and assisting scientists conducting research in some of the world’s
harshest climates [4]. They’ve even made their way inside our residences, where
they operate as both a source of entertainment and an indispensable tool for photographers. Many civil engineering applications like monitoring bridge safety, land
surveying, crack detection can also be carried out using UAVs [5].
Fig. 1 Different types of UAVs developed in recent times
Implementation of Machine Learning Techniques …
19
Machine learning (ML) is an artificial intelligence subset that splits data into
testing and training sets in order to construct prediction and forecasting models. Kneural networks, support vector machines, and Random forests are some of the
machine learning methods used for modelling and prediction [6]. By means of
adopting these ML algorithms in UAVs, optimal trajectories of motion become
possible. It allows the UAV can be implemented in different terrain and can be
used in different weather conditions [7]. The present work provides a comprehensive review of the scope, future prospects of Machine Learning algorithms from the
UAV applications. It discusses in detail the UAV classification, its market trends and
market values in the present times. The various machine learning algorithms and
their classification is also briefed. In addition, the adoption of these ML algorithms
in UAVs and their different applications are also explained.
2 Classification of UAV
Classification of UAVs can be based on a variety of performance parameters [8]. Wind
loading, speed, range, endurance, and weight are all critical factors that distinguish
various types of UAVs and serve as the foundation for useful categorization. It can
be a subsection depending upon the Strategic, Tactical, Size of the UAV, and Special
task [9]. Table 1 provides the detailed classification of UAV and their endurance,
flight altitude, range, and mass.
In addition to the above categorization, UAVs are categorized according to
measurements or parameters that include price, maximum take-off weight, engine
type, and pricing. Drones can be classified by to their range (short or long), pricing
(expensive or affordable), payloads (high or low), model complexity (complex or noncomplicated), number of blades (quadcopter, octocopter, multi-copter), and other
factors. Further, the UAVs are classified into multi-copter, unmanned helicopter, tiltwing, and fixed-wing [10]. These types of classifications provided by the researchers
are listed in Table 2.
3 Unmanned Aerial Vehicle (UAV) Market Trends
and Values
In the last six years, the market for unmanned aerial vehicles (UAVs) has exploded.
Figure 2 depicts commercial UAV revenue from 2015 to 2021, as well as predicted
revenue and forecasted up to 2030 [11], while Fig. 3 depicts UAV market values in
various sectors.
The worth of enterprises and labour in various industrial sectors that can benefit
from drone utilization demonstrates the significance of various UAV applications.
In the presented industrial sectors, the overall value of drone-powered solutions
20
E. Fantin Irudaya Raj
Table 1 Detailed classification of UAV
Depend upon the strategic (TBDa )
Types
Mass (kg)
Range (km)
Flight altitude (m)
Endurance (h)
High altitude long
endurance
2500–5000
>2000
20,000
22–44
Stratospheric
>3000
>2200
>25,000
>44
Exo-stratospheric
TBDa
TBDa
>35,000
TBDa
Medium altitude long
endurance
1000–1500
>600
4000
24–48
Long altitude long
endurance
15–25
>600
3000
>24
Low altitude deep
penetration
250–2000
>250
100–9000
0.5–1
Medium range
endurance
500–1500
500
8000
10–18
Medium range
150–500
70–200
4000
6–10
Short range
50–250
30–70
3000
3–6
Close range
25–150
10–30
2500
2–4
Mini
<25
<10
150
<2
Micro
<5
<10
250
1
Decoy
150–250
0–500
50–5000
>4
Lethal
TBDa
300
4000
3–4
Unmanned combat
>1000
2000
14,000
2
Depend upon the tactical
Depend upon the size
Depend upon the
special Task
a
TBD track before detect
Table 2 Types of UAV and their merits and demerits
Types
Multi-copter
Unmanned
helicopter
Merits
Easy to launch,
High payloads and Combination of
lightweight, and low vertical take-off and VTOL and
cost
landing (VTOL) are fixed-wing
both possible
Endurance over a
long distance
Demerits
Due to its low
weight and limited
payloads, it is
susceptible to wind
Horizontal take-off
necessitates a large
amount of area
Expensive and
high-maintenance
requirements
Tilt-wing
Expensive
technology
Fixed-wing
Implementation of Machine Learning Techniques …
21
Fig. 2 Commercial UAV Revenue from 2015 to 2021 and forecasted till 2025
Fig. 3 Market value of UAV in different sectors in 2021
exceeds USD 127 billion. Infrastructure (railways, roads, energy, and oil and gas)
has the biggest potential for drone use, with a total value of USD 45 billion [12].
Insurance, entertainment and media, telecommunications, agriculture, security, and
mining are all industries that use UAVs significantly.
4 Machine Learning Techniques for UAV Applications
The performance of the UAV is wholly dependent upon the algorithm which it is
adopted and the sensory devices are carried out. To enhance the performance of
UAV, the ML algorithms have been adopted in various stages like maneuvering of
UAVs, obstacle identification, communication between the ground station and the
UAVs, payload management, and Image analysis. By means of adopting a suitable
ML algorithm for a specific application will enhance the performance overall performance of UAVs [13]. This will also improve the applications of UAVs in various
22
E. Fantin Irudaya Raj
Fig. 4 Different machine learning algorithms
sectors. There are a number of research works carried out towards the application
of ML integrated UAV in computer/wireless networks, agriculture, smart city, forest
conservation, disaster management, and military.
Artificial Intelligence includes machine learning as a subset. UAVs can benefit
from machine learning to optimize differentiable parameters and operations. ML
algorithms, unlike software that has been manually programmed and executes tasks
with specific instructions, are designed in such a way that they may learn and improve
over time when exposed to new data. Figure 4 depicts the detailed classification of
different machine learning algorithms used in UAV applications [14]. If the data
signal obtained from the sensors and environment is huge means, we need to go
for the dimension reduction. For this purpose, we are using unsupervised learning
algorithms like Principal Component analysis, singular value decomposition, and so
on. If the responses are enabled and true, we can choose the Unsupervised learningbased clustering algorithms. Gaussian Mixture model, K-means, and K-nodes are
examples of this type of ML algorithms [15]. If the predicted value is numeric, we can
go with the supervised learning-based regression algorithms like Linear regression,
Neural network, Decision tree, etc. If the predicted value is not numeric means,
we can prefer the supervised learning-based classification algorithms like logistic
regression, Random Forest, Support Vector Machine (SVM), and so on [16].
In the following subsections, a few of the important Machine learning algorithms
used in UAV applications are explained.
Implementation of Machine Learning Techniques …
23
Fig. 5 Simple linear
regression
4.1 Linear Regression
It’s a technique for estimating real values from a continuous variable. We can establish
a link between the dependent and independent variables by fitting the best line. The
best fit line is a regression line, which is represented by a linear Eq. 1.
y = (a ∗ x) + b
(1)
where, a = Slope, b = Intercept, x = Independent Variable, y = Dependent Variable.
These a and b are coefficients, which are derived based on minimizing the sum of
squared difference of data between data points and the regression line [17]. Figure 5
represents the simple linear regression.
4.2 Logistic Regression
It’s a classification algorithm, not a regression one. It’s a method for estimating
discrete values from a group of independent variables. In simple terms, it fits data
to a logit function to forecast the probability of an event occurring. As a result, it’s
also called logit regression. Because it forecasts probability, the output values range
from 0 to 1 [18]. Figure 6 shows the simple logistic regression.
24
E. Fantin Irudaya Raj
Fig. 6 Simple logistic regression
4.3 Decision Tree (DT)
It’s a supervised learning algorithm that’s commonly used to solve classification
difficulties. It works for both categorical and continuous dependent variables, which
is surprising. We divide the population into two or more homogenous sets using this
approach. This is done to create as many separate groups as feasible based on the most
important attributes/independent variables [19]. It employs a variety of approaches,
including Gini, Information Gain, Chi-square, and entropy, to divide the population
into separate heterogeneous groups. Figure 7 shows the simple Decision Tree (DT).
4.4 Random Forest (RF)
A random forest is a collection of multiple description-based classifiers made up
of numerous DTs, just like a forest [20]. Overfitting of something like the training
samples is common with deep DTs, resulting in a massive fluctuation in classification outcomes for tiny changes in the input data. This algorithm seems acutely
susceptible to their training data, rendering them vulnerable to errors while dealing
with the test dataset. Different sections of the training dataset are used to train the
various DTs of an RF. Each DT of the forest must transmit down the input vector of
the new sample size to categorize it. The classification conclusion is subsequently
determined by each DT considering a separate section of the input vector. After that,
the forest determines whether to employ the class with the most “votes” (for distinct
Implementation of Machine Learning Techniques …
25
Fig. 7 Decision tree
Fig. 8 Pictorial representation of random forest with multiple decision trees
categorization decisions) or the total among all forest trees (for numeric classification
outcome). The RF algorithm can minimize the issues produced by only evaluating
one DT for the same dataset because it takes into account the results of numerous
separate DTs. The RF method is illustrated in Fig. 8.
4.5 Support Vector Machine (SVM)
The SVM algorithm is capable of classifying both linearity and non-linearity input.
It begins by mapping every data point through an n-dimensional attribute vector,
whereby ‘n’ indicates the total number of attributes.
26
E. Fantin Irudaya Raj
Fig. 9 Illustration of SVM
work. Here hyperplane
maximizes the separation of
two different classes (‘star’
and ‘circle’)
The hyperplane that splits these data elements into two groups is then identified,
with the minimum separation for both categories is maximized and classifying errors
minimized [21]. The minimal proximity for a grouped class is defined by the length of
both the selection hyperplane and the class’s adjacent occurrence. In more technical
terms, each feature vector is first plotted as a point in an ‘n’ dimensional space,
from each feature’s value becoming the value of a given coordinate. To complete the
categorization, we must first locate the hyperplane that separates the two categories
by the greatest margin. An SVM classifier is seen in Fig. 9 in a basic workflow with
a hyperplane.
5 Applications of Machine Learning Techniques in UAV
Numerous works reported in the literature about the applications of various machine
learning techniques in UAV based applications. UAVs with machine learning help
retrieve data via multi-dimensional mapping, develop infra, and to perform smart
farming. Artificial neural network (ANN) models are used to predict agricultural
yields [22], alleviate wetland diminishment for highway management [23], predict
evaporation [24], estimate plant water consumption [25], classify plants based on
their leaves [26], and simulate the rainfall-runoff process, to name a few examples.
The biological process and structural modelling were carried out using UAV image
data [27].
The UAV can take multispectral photos, and the user can adjust the image resolution by flying at various altitudes. Even though, without machine learning methods, it
is impossible to comprehend high-resolution photographs [28]. Random forest (RF)
is an image classification algorithm that makes use of bagging or bootstrap aggregation [29]. The RF technique [30] can be used to provide spectral estimates. In terms of
performance, random forest-based modelling outperforms support vector machines
and artificial neural networks [31]. In quantitative remote sensing data, extreme
Implementation of Machine Learning Techniques …
27
learning machine methods are employed to solve regression and classification issues
[32, 33].
Several researchers have successfully deployed unmanned aerial vehicles (UAVs)
in computer/wireless networks. UAV aided wireless networks were managed and
controlled using the Echo State Network (ESN) and Multi-agent Q-learning [34].
For analyzing UAV behaviour in various controlling situations, a deep convolutional
neural network was used [35]. In the UAV-based network, the liquid state machine
(LSM) was used as a machine learning method for resource management [36].
The use of combining machine learning and UAV photography in water management and irrigation [37], predicting soil moisture content [38], has been demonstrated. Crop classification and vegetation yield prediction have also been made
using multiple linear regression, support vector machines, and random forests [39–
42]. UAVs are increasingly being used in smart cities and the military for a number
of objectives. A graffiti clean-up system was built using machine learning algorithms
and the UAV platform [43]. A machine learning approach was used to classify, track,
and detect flying objects, whether manned or unmanned [44]. UAV detection and
identification were carried out using the RF signals provided by the UAV controller
[45]. Aissa et al. [46] used Markov models, Naive Bayes, and CNN with UAV to
classify UAV kinds based on observed communication signals among the consumer
UAV controller and the ground station.
Additional applications are indicated by various academics, including creating
statistical records for the mining industry, geology, and wildlife. For animal detection,
machine learning techniques with UAVs were utilized [47]. Using machine-learning
approaches, the Multiclass Relevance Vector Machine, Gradient Tree Boost, random
forest, k-nearest neighbour, and Support vector machine was used to classify geological extraction models based on the surface feature detection [48]. Table 3 summarizes
the most current works on UAVs and Machine Learning that have been published in
the literature.
6 Summary and Discussion
When machine learning and unmanned aerial vehicles (UAVs) are combined, image
classification and object recognition become more efficient, accurate, and precise.
Based on the data collected throughout this investigation, Fig. 10 demonstrates the
results of merging machine learning with UAV research. The United States is the
unchallenged leader in the disciplines of machine learning and unmanned aerial
vehicles. It is followed by Asia Pacific countries, which account for 38.9% of the
study. Africa and Latin America are still lagging behind in terms of technology, with
the least amount of machine learning and UAV research. The Asia-Pacific region has
an advantage over Europe in terms of machine and robotics innovation, with Korea
and Japan leading the way.
Nonetheless, the use of UAVs for product delivery, precision agriculture,
surveying and mapping, imaging, aerial remote sensing, and monitoring is expected
28
E. Fantin Irudaya Raj
Table 3 A summary of the most recent research on UAVs and machine learning
Sl. No.
Area of application
Machine learning algorithms adopted
1
Military
Support vector machine, random forests,
and K-nearest neighbours
2
Wild life
Support vector machine, logistic regression
3
Agriculture
Support vector machine, random forests,
and extreme learning machine, artificial
neural network, multi linear regression and
random forests
4
Smart cities
Convolutional neural network
5
Computer and wireless network
Logistic regression and linear regression
6
Traffic management
Convolutional neural net
7
Geology and mining
K-Nearest neighbours, support vector
machine
8
Identification and detection
Markov models, naïve Bayes, linear and
logistic regression
9
Disaster management
Deep convolutional neural network
10
Damage detection and land surveying
Decision trees and deep convolutional
neural network
Fig. 10 Machine learning
and UAV based research by
region wise
to increase [49]. The expected UAV market in various parts of the world by 2025
is depicted in Fig. 11. In several sectors, the United States could have a competitive advantage over any region by 2025 when it comes to the use of UAVs. The
Asia Pacific region might well continue to account for the greatest share of the UAV
industry. Europe (with a 9% share) and the Middle East (with a 7% share) are the
other players (8% share).
Figure 12 summarizes the research conducted on machine learning and unmanned
aerial vehicles in several fields over the last five years. Since 2017, there has been a
surge in UAV and machine learning research for military and smart city applications.
In 2017, it was only 20%, but by 2021, it would have risen to 42%. At the same
Implementation of Machine Learning Techniques …
29
Fig. 11 UAV market forecasting of the year 2025
Fig. 12 Machine learning and UAV research interact in various fields
time, machine learning and unmanned aerial vehicle (UAV) research in agriculture
continue to advance at a breakneck speed. According to a review of the literature,
the use of UAVs and machine learning in wireless/computer networks commenced
late in 2017 and continued into 2018. After all, it gained popularity throughout 2019
and 2020.
Additionally, Fig. 13 indicates the rise in popularity of machine learning algorithms throughout the last decade in a multitude of sectors. The association for
machine learning and unmanned aerial vehicles has made extensive use of a range
of algorithms. The random forest algorithm is the most popular of all algorithms.
Due to its capacity to deal with noise in the data, it is the most frequently employed
algorithm. With a 21% share, support vector machines are the second most prevalent.
30
E. Fantin Irudaya Raj
Fig. 13 Machine learning algorithms utilized in UAV
Convolutional neural networks and k-nearest neighbours are widely used, as indicated by their respective market shares of 16 and 11%. Algorithms such as multi-agent
learning, liquid state, and Naive Bayes have been used sparingly [50].
However, using UAVs and machine learning has a number of shortcomings in
the present time. New ML algorithms and user interfaces need to be created and
designed. For the drone manufacturing industry, the design of UAV-assisted wireless
networks is a source of concern [51]. One of the deployment issues is the intellectual
and workload inflicted on UAV operators and their companies. Similarly, controlling
UAV operations necessitated a larger workforce than typical aircraft. UAVs outfitted
with infrared sensors and facial recognition software are viewed as compromising
individuals’ privacy. Consumer UAVs can occasionally go unnoticed when using
conventional radar, vision, and sound detection technologies [52, 53]. In civil society,
the use of UAVs for surveillance is a hot topic. Huge training data sets are a common
stumbling block for machine learning algorithms, making them both costly and
time-consuming.
7 Conclusion
The UAV’s portability, lightweight, excellent resolution, small size, and ability to fly
at low altitude are all advantages. Machine learning and unmanned aerial vehicles
(UAVs) offer immense potential in scientific research. The chapter presented here
discussed and investigated the numerous machine learning techniques used in UAVs
in recent years, as well as thorough classification and market predictions for UAVs.
The study on combining UAVs and machine learning is still in its early stages. The
Implementation of Machine Learning Techniques …
31
current study discovered that research in this field is frequent, with the majority
of it focused on agriculture, smart cities, computer/wireless networks, statistical
analysis of wild animals, the military, and mining. The support vector machine and
random forest algorithms have been used in a number of ways in UAVs. The United
States and Asia-Pacific region account for the lion’s share of UAV utilization and
research. The rise of unregistered commercial UAVs has raised security and privacy
issues. Researchers intend to develop modern technologies and machine learning
algorithms for identifying and recognizing unregistered consumer UAVs. Machine
learning algorithms for recognizing objects in UAV and satellite imagery will be
developed in the future. These all will make the UAV more user-friendly, safe, more
reliable and cost-effective.
References
1. K.P. Valavanis, G.J. Vachtsevanos (eds.), Handbook of unmanned aerial vehicles, vol. 2077
(Springer, Netherlands, Dordrecht, 2015)
2. P.K.R. Maddikunta, S. Hakak, M. Alazab, S. Bhattacharya, T.R Gadekallu, W.Z. Khan, Q.V. Pham, Unmanned aerial vehicles in smart agriculture: applications, requirements, and
challenges, IEEE Sens. J. (2021)
3. J. Kim, S. Kim, C. Ju, H. Il Son, Unmanned aerial vehicles in agriculture: a review of perspective
of platform, control, and applications, IEEE Access 7, 105100–105115 (2019)
4. A. Otto, N. Agatz, J. Campbell, B. Golden, E. Pesch, Optimization approaches for civil applications of unmanned aerial vehicles (UAVs) or aerial drones: a survey. Networks 72(4), 411–458
(2018)
5. V. Hoskere, J.-W. Park, H. Yoon, B.F. Jr Spencer, Vision-based modal survey of civil
infrastructure using unmanned aerial vehicles, J. Struct. Eng. 145(7) 04019062 (2019)
6. Dr. Jain, Dr M.D. Sangale, E. Raj, A pilot survey of machine learning techniques in smart grid
operations of power systems, Europ. J. Mole. Clin. Med. 7(7), 203–210 (2020)
7. P.S. Bithas, E.T. Michailidis, N. Nomikos, D. Vouyioukas, A.G. Kanatas, A survey on machinelearning techniques for UAV-based communications. Sensors 19(23), 5170 (2019)
8. M. Arjomandi, S. Agostino, M. Mammone, M. Nelson, T. Zhou, Classification of unmanned
aerial vehicles, Report for Mechanical Engineering class, University of Adelaide, Adelaide,
Australia (2006)
9. Y. Alghamdi, A. Munir, H.M. La, Architecture, classification, and applications of contemporary
unmanned aerial vehicles, in IEEE Consumer Electronics Magazine (2021)
10. C. Lee, S. Kim, B. Chu, A survey: flight mechanism and mechanical structure of the UAV, Int.
J. Precis. Eng. Manuf. 1–25 (2021)
11. A. Erceg, B.Č. Erceg, A. Vasilj, Unmanned aircraft systems in logistics–legal regulation and
worldwide examples toward use in Croatia, Bus. Logistics Mod. Manage. (2017)
12. A. Erceg, Z. Kilic, The use of unmanned aerial vehicles: a comparison of Turkey and Croatia, in
Unmanned Aerial Vehicles in Civilian Logistics and Supply Chain Management, (IGI Global,
2019), pp. 64–85
13. L.N. Habibi, T. Watanabe, T. Matsui, T.S.T. Tanaka, machine learning techniques to predict
soybean plant density using UAV and satellite-based remote sensing, Remote Sens. 13(13),
2548 (2021)
14. E.F.I Raj, V. Kamaraj, Neural network based control for switched reluctance motor drive, in
2013 IEEE International Conference on Emerging Trends in Computing, Communication and
Nanotechnology (ICECCN) (IEEE, 2013), pp. 678–682
32
E. Fantin Irudaya Raj
15. A.C. Sijini, E. Fantin, L. Prakash Ranjit, Switched reluctance motor for hybrid electric vehicle,
Middle-East J. Sci. Res. 24(3), 734–739 (2016)
16. E.F.I. Raj, M. Balaji, Analysis and classification of faults in switched reluctance motors using
deep learning neural networks, Arab. J. Sci. Eng. 46(2), 1313–1332 (2021)
17. D.C. Montgomery, E.A. Peck, G. Geoffrey Vining, Introduction to linear regression analysis.
(John Wiley & Sons, 2021)
18. G. Ch, S. Jana, S. Majji, P. Kuncha, A. Tigadi, Diagnosis of COVID-19 using 3D CT scans
and vaccination for COVID-19, World J. Eng. (2021)
19. P. Agarwal, M.A. Ch, D.S. Kharate, E. Fantin Irudaya Raj, S. Balamuralitharan, Parameter estimation of COVID-19 Second Wave BHRP transmission model by using principle component
analysis, Annal. Rom. Soc. Cell Biol. 446–457, (2021)
20. Y. Chen, W. Zheng, W. Li, Y. Huang, Large group activity security risk assessment and risk
early warning based on random forest algorithm. Pattern Recogn. Lett. 144, 1–5 (2021)
21. M. Deivakani, S.V. Sudheer Kumar, N.U. Kumar, E. Fantin Irudaya Raj, V. Ramakrishna, VLSI
implementation of discrete cosine transform approximation recursive algorithm, J. Phys. Conf
Ser. 1817(1), 012017, IOP Publishing, (2021)
22. S. Khairunniza-Bejo, S. Mustaffha, W.I.W. Ismail, Application of artificial neural network in
predicting crop yield: a review. J. Food Sci. Eng. 4(1), 1 (2014)
23. B. Zaman, M. Mckee, A. Jensen, UAV, machine learning, and GIS for Wetland mitigation in
Southwestern Utah, USA, (2017)
24. K.P. Sudheer, A.K. Gosain, D. Mohana Rangan, S.M. Saheb, Modelling evaporation using an
artificial neural network algorithm, Hydrol. Processes 16(16), 3189–3202 (2002)
25. D.M. Qiao, H.B. Shi, H.B. Pang, X.B. Qi, F. Plauborg, Estimating plant root water uptake using
a neural network approach. Agric. Water Manag. 98(2), 251–260 (2010)
26. S.G. Wu, F.S. Bao, E.Y. Xu, Y.-X. Wang, Y.-F. Chang, Q.-L. Xiang, A leaf recognition algorithm for plant classification using probabilistic neural network, in 2007 IEEE international
symposium on signal processing and information technology (IEEE, 2007), pp. 11–16
27. I. Ahmad, M. Basheri, M. Javed Iqbal, A. Rahim, Performance comparison of support vector
machine, random forest, and extreme learning machine for intrusion detection, IEEE Access
6, 33789–33795 (2018)
28. V. Gampala, M. Sunil Kumar, C. Sushama, E. Fantin Irudaya Raj, Deep learning based image
processing approaches for image Deblurring, Mater. Today Proc. (2020). https://doi.org/10.
1016/j.matpr.2020.11.076
29. P.O. Gislason, J.A. Benediktsson, J.R. Sveinsson, Random forests for land cover classification.
Pattern Recogn. Lett. 27(4), 294–300 (2006)
30. K. Priyadarsini, E. Fantin Irudaya Raj, A. Yasmine Begum, V. Shanmugasundaram, Comparing
DevOps procedures from the context of a systems engineer, Mater. Today Proc. (2020). https://
doi.org/10.1016/j.matpr.2020.09.624
31. R.K. Douglas, S. Nawar, M. Carmen Alamar, A.M. Mouazen, F. Coulon, Rapid prediction of
total petroleum hydrocarbons concentration in contaminated soil using vis-NIR spectroscopy
and regression techniques, Sci. Total Environ. 616, 147–155 (2018)
32. V. Khosravi, F.D. Ardejani, S. Yousefi, A. Aryafar, Monitoring soil lead and zinc contents via
combination of spectroscopy with extreme learning machine and other data mining methods,
Geoderma 318, 29–41 (2018)
33. J. Peng, A. Biswas, Q. Jiang, R. Zhao, J. Hu, B. Hu, Z. Shi, Estimating soil salinity from remote
sensing and terrain data in southern Xinjiang Province, China, Geoderma 337, 1309–1319
(2019)
34. X. Liu, Y. Liu, Y. Chen, L. Hanzo, Trajectory design and power control for multi-UAV assisted
wireless networks: a machine learning approach. IEEE Trans. Veh. Technol. 68(8), 7957–7969
(2019)
35. Y.S. Mandloi, Y. Inada, machine learning approach for drone perception and control, in International Conference on Engineering Applications of Neural Networks (Springer, Cham, 2019),
pp. 424–431
Implementation of Machine Learning Techniques …
33
36. M. Chen, W. Saad, C. Yin, Liquid state machine learning for resource and cache management
in LTE-U unmanned aerial vehicle (UAV) networks. IEEE Trans. Wireless Commun. 18(3),
1504–1517 (2019)
37. M. Romero, Y. Luo, B. Su, S. Fuentes, Vineyard water status estimation using multispectral
imagery from an UAV platform and machine learning algorithms for irrigation scheduling
management. Comput. Electr. Agric. 147, 109–117 (2018)
38. X. Ge, J. Wang, J. Ding, X. Cao, Z. Zhang, J. Liu, Xi. Li, Combining UAV-based hyperspectral
imagery and machine learning algorithms for soil moisture content monitoring, PeerJ 7, e6926
(2019)
39. T. Suzuki, T. Tsuchiya, S. Suzuki, A. Yamaba, Vegetation classification using a small UAV
based on superpixel segmentation and machine learning, J Remote Sens. Soc. Japan 36(2),
59–71 (2016)
40. E. Fantin Irudaya Raj, M. Appadurai, Minimization of torque ripple and incremental of power
factor in switched reluctance motor drive, in Recent Trends in Communication and Intelligent
Systems: Proceedings of ICRTCIS 2020, (Springer Singapore, 2021), pp. 125–133
41. S. Jónsson, RGB and multispectral UAV image classification of agricultural fields using a
machine learning algorithm, Student thesis series INES (2019)
42. N. Viljanen, E. Honkavaara, R. Näsi, T. Hakala, O. Niemeläinen, J. Kaivosoja, A novel machine
learning method for estimating biomass of grass swards using a photogrammetric canopy height
model, images and vegetation indices captured by a drone. Agriculture 8(5), 70 (2018)
43. G.-H. Kwak, N.-W. Park, Impact of texture information on crop classification with machine
learning and UAV images. Appl. Sci. 9(4), 643 (2019)
44. D. Steininger, V. Widhalm, J. Simon, A. Kriegler, C. Sulzbachner, The aircraft context
dataset: understanding and optimizing data variability in aerial domains, in Proceedings of
the IEEE/CVF International Conference on Computer Vision, (2021), pp. 3823–3832
45. I. Nemer, T. Sheltami, I. Ahmad, A. Ul-Haque Yasar, M.A.R. Abdeen, RF-based UAV detection
and identification using hierarchical learning approach, Sensors 21(6), 1947 (2021)
46. S. Ben Aissa, A. Ben Letaifa, UAV Communications with machine learning: challenges,
applications and open issues, Arab. J. Sci. Eng. 1–21 (2021)
47. A.M. Dujon, D. Ierodiaconou, J.J. Geeson, J.P.Y. Arnould, B.M. Allan, K.A. Katselidis, G.
Schofield, Machine learning to detect marine animals in UAV imagery: effect of morphology,
spacing, behaviour and habitat, Remote Sens. Ecol. Conserv. (2021)
48. B. Krishna, G. Amuthavalli, D. StalinDavid, E.F. Raj, D. Saravanan, Certain investigation of
SARS-COVID-2-induced Kawasaki- like disease in Indian Youngsters. Annal. Rom. Soc. Cell
Biol. 1167–1182 (2021)
49. M. Yaqot, B.C. Menezes, Unmanned aerial vehicle (UAV) in precision agriculture: business
information technology towards farming as a service. in 2021 1st International Conference on
Emerging Smart Technologies and Applications (eSmarTA), (IEEE, 2021), pp. 1–7
50. T. Muezzinoglu, M. Karakose, An intelligent human-unmanned aerial vehicle interaction
approach in real time based on machine learning using wearable gloves. Sensors 21(5), 1766
(2021)
51. M. Fourati, B. Najeh, A. Idriss, Blockchain Towards Secure UAV-Based Systems, in Enabling
Blockchain Technology for Secure Networking and Communications, (IGI Global, 2021),
pp. 149–174
52. E. Raj, F. Irudaya, M. Appadurai, The Hybrid Electric Vehicle (HEV)—An Overview, Emerging
Solutions for e-Mobility and Smart Grids, (2021), pp. 25–36
53. I. Colomina, P. Molina, Unmanned aerial systems for photogrammetry and remote sensing: a
review. ISPRS J. Photogramm. Remote. Sens. 92, 79–97 (2014)
Machine Learning Techniques for UAV
Trajectory Optimization—A Survey
Sasikumar Rajendran, Karthik Kandha Samy, Jeganathan Chinnathevar,
and Deva Priya Sethuraj
Abstract Unmanned Aerial Vehicles (UAVs), generally known as drone are utilized
in various genuine applications like payload conveyance, traffic observing, moving
articles in apparently risky climate, and observation. When utilized in mechanical
scenery, UAVs can shape a center piece of modern automation alongside IoT gadgets.
With regard to the traditional applications, their trend of getting in similar domains
is expected to have enhanced outputs, inspite of many new challenges. To meet out
the issues in this scenario, the Artificial Intelligence support is needed to play a vital
role when UAVs are applied. A deep study is made and presented in this work with
a keen study of all Machine Learning techniques utilized for UAV linked changes,
for example, channel displaying, asset the board, situating, and security.
Keywords Unmanned aerial vehicles · Machine learning · Neural network ·
Trajectory prediction · Drone · 5th generation wireless
1 Introduction
The basic goals needed in Fifth era (5G) and past trades are the growth of huge organizations with expounded throughput, in the scene of emerging trends in Internet of
Things (IoT) with varied settings. To meet out this, UAVs are expected to provide
fast responses in any jam-packed scenarios. These qualities have haggard in light of a
S. Rajendran (B)
K.Ramakrishnan College of Engineering, Samayapuram, Tamil Nadu 621112, India
K. K. Samy
M.Kumarasamy College of Engineering, Karur, Tamil Nadu 639113, India
e-mail: karthikk.cse@mkce.ac.in
J. Chinnathevar
iNurture Education Solutions Pvt. Ltd, Bengaluru, India
D. P. Sethuraj
SRM TRP Engineering College, Tiruchirappalli, Tamil Nadu 621105, India
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_3
35
36
S. Rajendran et al.
valid concern for the normalization bodies [1] and the learned society. The advancement of UAV in remote correspondence networks has given a few opportunities for
novel organization ideal models with expanded adaptability and execution. Nonetheless, the utilization of AI methods in thick remote organizations has permitted the
improvement of low-intricacy answers for a general organization streamlining from
the actual layer up to asset and organization the executives. The following subsections
reveal the current spaces of Machine Learning and UAVs.
1.1 What is UAV?
As the interest for wide broadband associations, generally joining, and boundless
access has made, non-regular affiliations (NTNs) can unequivocally keep up with the
grounded earthbound networks [2]. Amid the standard bits of NTNs, with respect to
the applications of IoT [3], the Low-stature stages mean to work with different ordinary inhabitants, business and legitimate missions, going from armed and defense
endeavours into redirection with broadcast exchanges. Basic specialist kind are negligible modernized planes utilized for brief timeframes permitting the swift relationship of multi-bounce correspondence spine in testing with no workforce included,
like public security, search and salvage missions, insight review, crisis trades in
post-disaster conditions or amazing occasions, visual perception, metropolitan traffic
reconnaissance [4, 5], accuracy agribusiness, and media traffic checking [6]. It is
critical about quantifiable investigation ascertains courses of action of UAVs [7, 8].
1.2 Machine Learning with Artificial Intelligence
Man-made understanding has been viewed as the examination of preparation
machines for performing undertakings of human. Different purposes which manmade intelligence connects include modernized means of transportation, talk assertion, machine interpretation, and really remote exchanges. Additionally, a particular
subset of man-made intelligence is the philosophies that are utilized for preparing
machines in how to recognize, which starts another structure known as Machine
Learning which can give plans with conditions in endless contraptions at the same
time expect that induction should the association\’s assets as in IoT correspondences.
This logic, sharps the bosses ought to be acted in the whole relationship to conform
to the unmistakable referencing necessities of this wise sort of associations. The
expansion is to adaptively and continually deal with the affiliation assets ideally. As
needs are, ML estimations have been proposed as a useful strategy for going toward
this heap of incoherent troubles coming from the IoT natural framework.
Machine Learning Techniques for UAV Trajectory Optimization …
37
2 Survey Works
By far most diagrams either centre on ML for various far off affiliation conditions or
expect the coordination in affiliations, examining its prospect potential. Regardless,
this review is based on simulated intelligence for automated applications. In such
conditions, they can work on association and protection of multi-robot packs when
sufficiently playing out some undertakings. Reenacted knowledge systems for flight
mechanical development unite (I) Artificial Neural Networks for postponement, association and security improvement; (ii) iota swarm progress for picking UAV bearing;
(iii) Deep Learning for extra made availability; and (iv) Machine Learning for client
mentioning supposition. Bases on various correspondence frameworks including
space, air, and land areas, expressly Space-Air-Ground solidified affiliation were
examined. It is discovered that DL, basically CNNs, and various Deep Learning
designs and preparing frameworks will also encourage affiliation execution.
On the outline, an extensive game plan of simulated intelligence methods is
focused for Drone-set up correspondences without focusing an explicit simulated
intelligence/ML orders or UAV applications. Even more unequivocally, the going
with responsibilities is given: A total diagram of computer-based intelligence/ML
plans from every conceivable request will be introduced. An extensive degree of
Drone-overhauled far off correspondence issues, going from genuine layer and
asset the blockade perspectives to heading plan and taking care of is examined,
while distant precautions and open success functions are extensively talked about.
Untie concerns are recognized on behalf of the two frameworks organization and
precautionary vicinity, strengthening added assessment for the usage of simulated
intelligence methods.
Unique employments in computer-based intelligence game plans in Droneengaged correspondences are depicted in Fig. 1. Generally, the significance of
amending knowledge in distant correspondence networks has been spread out in
different works. The makers have seen that the reliably growing heterogeneity and
unpredictability of convenient associations have made checking and the leading
body of association parts steady. Additionally, ML licenses intentional pulling out
of significant processed data from adaptable normally recognizes connections which
are disapproved boggling in any capacity deduced by human subject matter experts.
Additionally, in far off associations, it facilitates the distant devices to adequately and
admirably screen present situation, exploiting adaptable data for planning purposes
to anticipate, acclimate headway of ecological components, together with remote
channel components and transportability plans [9].
2.1 Issues on Physical Layer
A wide extent of correspondence points [10] exists that can be worked on through ML
game plans, including (I) the headway of exact direct models in various conditions
38
S. Rajendran et al.
Fig. 1 Artificial intelligence/machine learning communications
and the mitigation of way disaster (PL) through the figure of the geology; (ii) the
treatment of genuine impedance from other UAVs and from ground centers’ with
ideal getting ready, using convey ability and customer association data; and (iii) the
plan of transmission limits towards achieving unequivocal execution targets. A wide
extent of correspondence perspectives exists that can be redesigned through ML game
plans, including (I) the progression of precise redirect models in various conditions
and the alleviation of way disaster (PL) through assumption for the topography;
(ii) the treatment of outrageous obstacle from other UAVs and from ground centers
with ideal planning, using flexibility and customer association data; and (iii) the
arrangement of transmission limits towards achieving express execution targets.
2.2 Channel—Modeling
Planning to focus on the utilization of Machine Learning on PL estimate for the
ethereal channels, Reference makers [11] came out with a distinctive ML-based PL
models for metropolitan conditions. Computations, involved sporadic boondocks
(RandF) and KNN. RandF is a gathering technique based on decision trees. With the
course of action, portrayals are assembled from every tree to pick the most estimate,
outlining an efficient classifier from innumerable fragile associated classifiers. It
executes directed study reliant upon a likeness compute, working out the distance
Machine Learning Techniques for UAV Trajectory Optimization …
39
between centers around an outline to deal with gathering or backslide issues. Then,
using shaft [12] following to deliver getting ready and testing data, the gauge precision
of the ML-based PL models was differentiated and the trial Stanford College Break
(SUI) and the COST231–Walfisch–Ikegami models. It was shown that the ML-based
models achieved better elevated PL figure by relying upon getting ready data related
to expansion distance, UAV stature, rise point, and LoS conditions.
Also, among the ML-based strategies, RandF offered the best assumption results
while the fundamental data was that of way detectable quality with spread distance
and stature point, moreover having basic effect. Another audit focused in on PL
conjecture for UAV networks talking with ground centers. Regardless, pulling out
from radio-repeat (RF) PL assumption, the scholars investigated in light of everything, PL and defer spread gauge in mmWave channels, using the RandF and KNN
ML computations. Furthermore, the issue of requiring immense enlightening files
for planning of the ML estimations is eased through move learning reliant upon
confined educational assortments and getting ready outcomes that were secured from
past getting ready stages. From the results, it is assumed that the proposed strategies offer diminished mean square botches when appeared differently in relation to
Okumura–Hata and COST 231–Hata models.
The considered ML computations contained the sporadic woodlands (RandF) and
KNN. RandF is a gathering procedure coming from decision trees. To perform game
plan, it assembles portrayals from every one of the decision trees and picks the most
notable gauge as its yield, outlining a strong classifier from endless delicate or weakly
associated classifiers. KNN performs managed learning subject to a likeness measure,
working out the distance between centers around an outline to deal with gathering
or backslide issues [13]. Then, using bar following to deliver getting ready and
testing data, the conjecture precision of the ML-based PL models was differentiated
and the exploratory Stanford College Interval (SUI) and the COST231–Walfisch–
Ikegami models. It was shown that the ML-based models achieved better aeronautical
PL estimates by relying upon getting ready data related to multiplication distance,
UAV tallness, rise point, and LoS conditions. Furthermore, among the ML-based
techniques, RandF offered the best assumption results while the primary data was that
of way detectable quality with spread distance and tallness point, moreover having
a basic effect. Another survey focused on PL figures for UAV networks talking with
ground centre points. Regardless, pulling out from radio-repeat (RF) PL assumption,
the authors assessed taking everything into account, PL and delay spread conjecture
in mmWave channels, using the RandF and KNN ML estimations. Also, the issue
of requiring gigantic instructive lists for getting ready of the ML computations is
reduced through move learning subject to confined enlightening assortments and
planning results that were obtained from past getting ready stages. From the results,
it is assumed that the proposed methods offer diminished mean square bungles when
appeared differently in relation to Okumura–Hata and COST 231–Hata models.
40
S. Rajendran et al.
2.3 Interference Management
Interference help in networks, in which both ground centre points and UAVs are
related by a cell. The target of the audit: To cultivate the way orchestrating plan for
the UAVs contemplating the impedance which is familiar with the ground centre
points and the next one is to restrict the inaction of information. Accordingly, a
DRL structure which is ESN-based for UAV (online) heading smoothing out is given
with a non-agreeable sport is arranged including the company methodologies with
conveying control stages freely. Furthermore, to vanquish the deterrent of understanding the whole ground geology, the UAVs take on the RL ESN-based computation with expectation of the base connecting components. Execution estimation
explained that the proposed loom can lessen the far off idleness in order to construct
a multifaceted nature with regard to lesser rate which simulates the customary most
restricted way scheme.
2.4 Configuration of Transmission Parameters
In huge various information different yield (MIMO) frameworks, precoding networks
are critical to work on the presentation of communication in taking advantage of CSI
data. A cross assortment precoding plan subject to a state of assembly show is given,
connecting elevated sufficiency in an immaterial expense. Appropriately, a strong
and energy-fit mix precoding planning is made, utilizing the ML cross-entropy (CE)
and the relative mistake assessment movement procedures. In the key stage, the cross
assortment plot abstractly conveys diverse expected fundamental precoders as per
the likelihood dissipating, surveying their reachable complete rates dependent upon
the conventional CE assessment. Then, at that point, the normal precoder is weighted
by the functional outright measures.
Concurrently, the comparative misstep among the invigorated assumption for
precoder and the next to ideal prospects is not really set in stone, while the general
bumble is being revived. In the last stage, a straightforward precoding system close
to the ideal probability is still up in the air. Reenactment results format that, regardless of the way that the proposed estimation achieves diminished absolute rate
appeared differently in relation to full progressed precoding, in light of damaged
system designing and structure display procure disasters. All the while, the cream
precoding plan is an achievable alternative for decreased energy use diverged from
full progressed precoding, radio wire assurance, and standard CE courses of action.
Table 1 includes different communications aspects with appropriate works and
relevant Machine Learning solutions.
Machine Learning Techniques for UAV Trajectory Optimization …
Table 1 Communication
barriers—machine learning
enhanced UAVs
41
Reference
Target
Solution based on
ML
Goudos et al.
RSS Prediction
ANN and DE LM
Learning
Wang et al.
Channel modeling Unsupervised
online
Chen et al.
Radio map
creation
Segmented
regression
Egi et al.
Prediction of PL
ANN with LiDAR
Zhang et al.
PL prediction
RandF with KNN
Yang et al.
PL prediction
RandF with KNN
Chen et al.
Competent
Dispersed
Challita et al.
Improvement of
Interference
ESN based RL
Athukoralage et al.
Interference
management
Based on regret
Zhang et al.
AMC - Optimized DL with CNN and
LSTM
Ren et al.
Procoding
Based on CE
3 Resource Management and Network Planning
ML procedures are normally applied to anticipate UAVs cell quality, which is related
to cell association. Even in addition unequivocally, one more unforeseen sporadic
ground point is planned to expect the finest portion of the cell in favour of a centre
point at region. The key thought is to take advantage of spatial affiliation which is
normally found in flying straits close by locations. For reviewing presentation of the
future advance, guaranteed 3GPP LTE multiplication limits is recognized for cell
model. Mathematical outcomes reveal the planned plot suggests enhanced accuracy
and further made execution when stood apart from different heuristics methods.
Recently, a brilliant viewpoint is given in distant affiliations [14], towards limiting
the front take inconvenience and lessening the transmission dormancy. Because of
their adaptability, these vehicles have been considered to show a key occupation
towards working on the introduction of store empowered affiliations. The producers
of reference took advantage of client-driven data identified with content mentioning
tasks and convenient plans for sending UAVs and for picking substance saving
money with support. Enduring with regard to dispersed enrolling gives with the secret
weapons for the specific content figures and client restrictions, ideal holding methodologies are also given. Singular client lead portrayed by the cloud into obvious models
through a conceptor-based ESN framework is empowered with exactness through
ML-based supposition. Thusly, the ESN-based technique works with the extraction
of the client Coalition, position, and stuff taking care of framework. The reenactments uncover that the UAV confer force can be adequately decreased showed up
42
S. Rajendran et al.
diversely according to an assessment without taking care of likewise as standard ESN
approaches, while QoE is essentially broadened wandered from geography without
UAVs.
4 Open Issues
Artificial Intelligence/Machine Learning—engaged and learning-driven remote
organizations will bring exceptional dynamic abilities and constant forecasts towards
altering 5G and past 5G organizations. Toward this path, the most recent headways
in the artificial intelligence/ML strategies released novel choices for the Unmanned
Aerial Vehicles based frameworks having prompted this chance of acknowledging
profoundly independent UAV tasks while upgrading presentation, defending the
safety, and alleviating individual deficiencies in complicated situations. The scope
of future exploration headings can be summed up as follows.
4.1 Implementation
Artificial Learning methodologies depend, on a fundamental level, on tremendous
and magnificent checked enlightening assortments to attain the best outcomes.
Though the information accessibility at present is monetarily potential and precisely
more moderate by the improvement of dispersed processing, information assembled
by sensors and association utensils are for the most part reliant upon incidents, overabundance, mislabeling, and class lopsidedness. Hence, the practicality of the readiness method is risky. At the present time, the Tensor Stream Light is used to effectively
utilize Machine Learning methods in addition to working with item affirmation on
uncertain data. Likewise, Keras a huge association application programming interface is prepared for executing on Tensor stream to enhance speedy implementation.
Likewise, the present astonishing multi-focus central planning unit (computer chip)
plans, GPUs, and far-reaching availability of libraries for DL consider fast, equivalent
data dealing with dynamically.
4.2 Issues in Physical Layer
Presently, a hole in securing information with broad estimation crusades exists [15].
In future, we can execute proving grounds complete genuine investigations in various
engendering regions, particularly in thick, metropolitan, high rise rich settings and
over ocean regions to approve the exactness of calculations, particularly among
situations in progressively altering conditions. e.g., vehicles, with expanded idleness, versatility prerequisites, when thinking about the impedance in proliferation
Machine Learning Techniques for UAV Trajectory Optimization …
43
region by examining in certifiable limitations, for example, the energy productivity
of the learned direction. By the by, these genuine issues regularly include very highdimensional nonstop status spaces, i.e., potentially activities, to formulate the relating
issues basically obstinate along with recent methodologies. In addition, estimated
information will be gained, Machine Learning methodologies be able to encourage
the latest advancements in control displaying.
4.3 Issues in Security and Privacy
A solid need exists for cutting edge techniques in doing combating truly making
assault types against the suffering idea of correspondence and the flourishing of
key foundations. Security stresses during the development of Unmanned Vehicles
structures join CP assaults focused on getting the development of Unmanned Vehicles
and harming or taking its heap. Likewise, for wise media structures depending upon
Unmanned Vehicles, upsetting the transmissions by conveying the characters may
have an authentic effect particularly in systems including a colossal quantity of
Unmanned Vehicles. Where, Unmanned Vehicle insistence with evaluation of got
sight and sound substance by the BS is required, accomplishing futile postponement.
Then, at that point, in cunning transportation frameworks, colossal quantities of
Unmanned Vehicles cooperate and trade information to assist with communicating
undertakings. Regardless, these crowds are powerless against aggressors that joint the
gathering and instil joke information that may accomplish bungles during progression
and conceivable naughtiness because of mishaps. To conform to the danger of illdisposed Machine Learning during harsh information transmission, bound together
learning is proposed, in which the game plan information of a particular Machine
Learning task is dealt with in a dissipated manner among Unmanned Vehicles with
gigantic number with regard the improvement issue managed in general.
5 Conclusion
A broad outline of the use of AI in networks with automated vehicles has been introduced. Additionally, the order of these strategies depends on the correspondence and
organization angle that is clearly presented. Similarly, all issues and solutions in each
huge area are also discussed, spreading out the meaning of drastic updates among
distant associations with requirements for extra movements reliant upon the use
of man-made intelligence in computerized raised vehicles correspondence associations. By and large, it’s fair that colossal proportions of data are open from different
sources; significant courses of action will acceptably uncover supportive connections
in heterogeneous data towards upgrading associations of UAV. In light of everything,
in circumstances where UAVs have been confined taking care of capacities or where
a serious cloud/dimness/edge system doesn’ exist to manage a ton of information; the
44
S. Rajendran et al.
strategy most likely won’t be basically functional. Thusly, considering the changing
dealing with limits, a necessity for self-sufficient, scattered harmonization, game
plans including some intricacy, close by an appraisal with the planning of limits
might be more appropriate. It is the inspiration driving why man-made intelligence
game plans, has taken on a couple of significant assessments.
References
1. 3GPP TR 36.777 V1.1.0: Study on enhanced LTE support for aerial vehicles. Technical
Specification Group Radio Access Network, (2018)
2. J. Liu, Y. Shi, Z.M. Fadlullah, N. Kato, Space-air-ground integrated network: a survey. IEEE
Commun. Surv. Tutor. 20, 2714–2741 (2018). https://doi.org/10.1109/COMST.2018.2841996
3. M.N. Hossein, T. Taleb, O. Arouk, Low-altitude unmanned aerial vehicles-based Internet of
Things services: comprehensive survey and future perspectives. IEEE Internet Things J. 3,
899–922 (2016). https://doi.org/10.1109/JIOT.2016.2612119
4. B. Li, Z. Fei, Y. Zhang, UAV communications for 5G and beyond: recent advances and future
trends. IEEE Internet Things J. 6, 2241–2263 (2019). https://doi.org/10.1109/JIOT.2018.288
7086
5. M. Mozaffari, W. Saad, M. Bennis, Y. Nam, M. Debbah, A tutorial on UAVs for wireless
networks: applications, challenges, and open problems. IEEE Commun. Surv. Tutor. 21, 2334–
2360 (2019). https://doi.org/10.1109/COMST.2019.2902862
6. A. Fotouhi, H. Qiang, M. Ding, M. Hassan, L.G. Giordano, A. Garcia-Rodriguez, J. Yuan,
Survey on UAV cellular communications: practical aspects, standardization advancements,
regulation, and security challenges. IEEE Commun. Surv. Tutor. (2019). https://doi.org/10.
1109/COMST.2019.2906228
7. D. Joshi, Commercial Unmanned Aerial Vehicle (UAV) market analysis: industry trends,
companies and what you should know (2017)
8. Federal Aviation Administration (FAA) Forecast fiscal years 2017–2038, (2018)
9. W. Choi, T.S. Shon, H.H. Choi, Y. Lee, Designing a novel unlicensed nomadic access
relay station in IEEE 802.16-based wireless access networks. in IEEE Vehicular Technology
Conference-VTC2007-Spring (VTC), (2007), pp. 2961–2965
10. Radio Technical Commission for Aeronautics (RTCA) SC-228, Minimum operational performance standards for unmanned aircraft systems (2013)
11. Radio Technical Commission for Aeronautics (RTCA) Drone Advisory Committee (DAC),
(2016) Available online
12. National Aeronautics and Space Administration (NASA) unmanned aircraft systems integration
in the national airspace system project (2016)
13. Microsoft Pegasus II: real-time internet of things experiment from the edge of space, (2016)
14. Facebook flying Aquila: Early lessons from the first full-scale test flight and the path ahead,
(2016)
15. S. Ananth, B. Wojtowicz, A. Cohen, N. Gulia, A. Bhattacharya, B. Fox, System design of the
physical layer for Loon’s high-altitude platform. EURASIP J. Wirel. Commun. Netw. 2019,
170 (2019). https://doi.org/10.1186/s13638-019-1461-x
Metaheuristic Algorithms for Integrated
Navigation Systems
Ali Mohammadi , Farid Sheikholeslam , and Mehdi Emami
Abstract This research evaluates novel and powerful metaheuristic optimization
approaches for designing integrated navigation systems. For this purpose, Inclined
Planes system Optimization (IPO) alongside its modified version called MIPO is used
for the first time. Implementations are done on an Inertial Navigation System (INS)
integrated with a Global Navigation Satellite System (GNSS). Noise covariance
matrices are considered as design variables and the sum of root-mean-squared errors
as an objective function in the form of a single-objective optimization problem.
Simulation results are reported in terms of all algorithmic and navigation performance
indicators. The overall assessment in comparison with two well-known competitors
of Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) indicates the
success of the proposed metaheuristic algorithms over the basic integrated navigation
problem.
Keywords Inclined planes system optimization · Integrated navigation · MEMS
IMU · Metaheuristic optimization · Soft computing
1 Introduction
In autonomous vehicles, navigation has one of two meanings: (a) estimating the
position, velocity, and attitude of the vehicle relative to a specific reference derived
from sensor observations; (b) design and carry out vehicle movements to reach the
desired location. Position, velocity, and attitude are called navigation states. The
A. Mohammadi (B) · F. Sheikholeslam
Department of Electrical and Computer Engineering, Isfahan University of Technology, Isfahan
84156-83111, Iran
e-mail: a.mohammadi98@pd.iut.ac.ir
F. Sheikholeslam
e-mail: sheikh@iut.ac.ir
M. Emami
Department of Electrical and Computer Engineering, Yazd University, Yazd 89158-18411, Iran
e-mail: m.emami@stu.yazd.ac.ir
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_4
45
46
A. Mohammadi et al.
term vehicle is also used for a moving body whose position and attitude must be
determined. Positioning methods are divided into three groups: inertial, satellite, and
integrated navigation. In inertial navigation, gyroscopes and accelerometers measure
rotation and specific force, respectively.
An Inertial Navigation System (INS) suffers from a sharp increase in error over
time, which is due to both the nature of the sensors and the type of mechanization. This can be solved by considering the required conditions, such as achieving
high accuracy sensors along with the use of appropriate and effective navigation
algorithms. To overcome the problems in INSs, however, aided navigation systems
such as Global Positioning System (GPS), also known as Global Navigation Satellite
System (GNSS) navigation systems, can be used alongside them. The main problem
in integrated navigation is using a properly integrated filter that provides a navigation
response in the shortest possible time and with the least error.
In integrated navigation, two models are used: one to model the state transition
process INS and the other to model GPS observation. In such models, the uncertainties in INS and GPS sensors are modeled with two noise sources, called process
noise and measurement noise, respectively (which can generally be non-Gaussian
and non-white). Based on these two models, and using the correction and updating
equations, the states estimated by the INS are corrected based on GPS observations.
There are several significant challenges in designing and implementing navigation
systems to achieve this goal. One of them is accurate modeling of the INS transition process and GPS observation. Therefore, if linearized models of the system and
Gaussian white noise are used, since the INS is essentially a nonlinear system with
non-Gaussian and non-white noise, the accuracy of the navigation system will be
reduced. Navigation algorithms based on the Linear Kalman Filter (LKF) and the
Extended Kalman Filter (EKF) use linear error state models. Despite the advantages
of Micro-Electro-Mechanical System (MEMS) inertial sensors (such as low cost,
small size, and low power consumption), as one of the most widely used sensors
in the field of inertial navigation, KF-based methods when using MEMS Inertial
Measurement Units (IMUs), from divergence during GPS outages, which result from
linearization process approximations and undesirable system modeling, are also not
safe. Kalman navigation filter as the core of the navigation system, especially integrated navigation, is an optimal estimation tool that provides a sequential recursive
algorithm for estimating system states [1].
In various studies, nonlinear methods have been proposed to compensate KF
defects. The use of other estimating filters such as Unscented Kalman Filter (UKF)
and Particle Filter (PF) and other theoretical methods to solve navigation problems
have also been pursued in them. Consequently, they have accepted the high volume of
calculations and theoretical considerations, and operational assumptions to achieve
the desired answers. In contrast, some studies have achieved intelligent and optimal
navigation systems by focusing on integrated navigation and exploiting the unique
potential of artificial intelligence-based soft computing approaches such as artificial
neural networks, fuzzy logic, deep and reinforcement learning, and evolutionary or
metaheuristic optimization algorithms [2–16]. Table 1 lists and reviews some of the
related research.
Metaheuristic Algorithms for Integrated Navigation Systems
47
Table 1 Recent related research
Research
Application
Method
Achievements
Date
Sathiya and
Chinnadurai [2]
Trajectory
planning of
mobile robot
Heterogeneous
Multi-Objective
Differential Evolution
(HMODE) and elitist
Non-dominated
Sorting Genetic
Algorithm (NSGA-II)
Optimal trajectory
planning of
differential-driven
Wheeled Mobile
Robot (WMR)
2019
Jalali et al. [3]
Autonomous
Training Multilayer
robot navigation Perceptron (MLP)
networks using six
Evolutionary
Algorithms (EAs),
Multi-Verse Optimizer
(MVO), Moth-Flame
Optimization (MFO),
PSO, Cuckoo Search
(CS), Grey Wolf
Optimizer (GWO), and
Bat Algorithm (BA),
for solving a
classification task
Robust
performance of
MVO for both
evaluation metrics
[Accuracy and
Area Under Curve
(AUC)], the
convergence
profile, and the
t-test results
2019
Liu et al. [4]
3D robot
navigation
Deep reinforcement
learning
Set up a robot
navigation
environment in
various dense
pedestrian
environments
2020
Cong et al. [5]
The
MEMS-based
GNSS/INS land
vehicle
navigation
system
Enhancing the
Signal-to-Noise Ratio
(SNR) of MEMS-INS
raw measurements
utilizing a hybrid
denoising algorithm
with wavelet transform
and Support Vector
Machine (SVM); and
improving the
positioning accuracy
by a SVM-based data
fusion approach
Effectively
2020
eliminate the
stochastic errors of
MEMS-IMU, and
significantly
improve the overall
positioning
accuracy
Yi et al. [6]
Path planning
navigation for
Uninhabited
Combat Air
Vehicles
(UCAV)
A Quantum Monarch
Butterfly Optimization
(QMBO)
Finding the much
shorter path by
QMBO than MBO
2020
(continued)
48
A. Mohammadi et al.
Table 1 (continued)
Research
Application
Method
Bellemare et al. [7]
Autonomous
navigation of
stratospheric
balloons
Reinforcement learning The proposed
2020
controller
outperforms
Loon’s previous
algorithm and is
robust to the
natural diversity in
stratospheric
winds; and the
reinforcement
learning is an
effective solution
to real-world
autonomous
control problems in
which neither
conventional
methods nor
human intervention
suffice
Achievements
Date
Bitar and Gavrilov [8]
Integrated
MEMS-based
INS/GNSS
UKF and Nonlinear
Autoregressive neural
networks with external
inputs (NARX)
(namely NARX aided
UKF)
The improvement
in position
accuracy is about
85–90%, in
velocity about
65–75%, and in
attitude about
40–65%
2020
Wang et al. [9]
Strapdown
Inertial
Navigation
System (SINS)
and GNSS
integrated
navigation
system
Multiple fading factor
Square root Cubature
Kalman Filter
(MSCKF) and
Generalized Dynamic
Fuzzy NN Model
Based on MSCKF
(MSCKF-GDFNN)
Reducing the
position errors in
latitude and
longitude by
85.00%, 89.71%,
and the velocity
errors in east and
north by 94.57%,
83.11%
2021
Gul et al. [10]
Path planning
for autonomous
guided robot
Grey Wolf
Optimizer-Particle
Swarm Optimization
(PSO–GWO)
algorithm with
evolutionary
programming as a
multi-objective path
planning algorithm
A more feasible
path with a short
distance and
overcoming the
Shortcomings of
other conventional
techniques
2021
(continued)
Metaheuristic Algorithms for Integrated Navigation Systems
49
Table 1 (continued)
Research
Application
Method
Achievements
Zieliński and
Markowska-Kaczmar
[11]
Vision-based
3D robotic
navigation
Deep reinforcement
learning
The ability to avoid 2021
collisions and find
the target object
quickly
Date
Gao et al. [12]
Inertial
navigation
system
A gravity
compensation method
based on multilayer
feedforward neural
network
Improving the
2021
radial position
error performance
more than 31.43%,
and ensuring the
real-time
performance of
gravity
compensation
Wen et al. [13]
Multi-robot
path-planning
for autonomous
navigation
Proposing a novel meta
reinforcement learning
framework based on
transfer learning, and
providing a dynamic
Proximal Policy
Optimization with
Covariance Matrix
Adaptation
evolutionary strategies
(dynamic-PPO-CMA)
to extend original PPO
algorithm
Faster convergence 2021
rate and arrive the
destination more
quickly
Bitar and Gavrilov
[14]
Integrated
INS/GNSS
system
Combining UKF and
Nonlinear
Autoregressive neural
networks with external
inputs (NARX)
Improving the
accuracy during
GNSS outages
2021
Yan et al. [15]
SINS/GPS
integrated
navigation
system
A deep neural network
(DNN)-based
Real-Time Sequence
Analyzer (RTSA) and
covariance matching
algorithm hybrid
adaptive nonlinear
filter for solving the
estimation of dynamic
state, time-varying
noise covariances Q
and R of SINS and GPS
Possessing high
generalization
capability and
accurately label
multi-frequency
compound
vibrations, and
showing better
accuracy,
robustness, and
flexibility
2021
(continued)
50
A. Mohammadi et al.
Table 1 (continued)
Research
Application
Method
Achievements
Date
Wu [16]
Aircraft motion
planning
A survey on
population-based
meta-heuristic
algorithms for solving
AMP problems
Providing some
suggestions on
how to select
appropriate
population-based
meta-heuristic
algorithms for a
particular AMP
problem
2021
In the present study, in line with many similar studies, emphasis has been placed
on using new approaches, including KF and AI hybrid systems, to complement each
other. Using a KF as the main integration filter and adjusting its parameters can
improve the performance of an INS [17]. Such methods are known as Adaptive
Kalman Filters (AKFs). The main methodology in such research is a method and
algorithm that can intelligently and completely optimally estimate the control values
of the algorithm and provide the appropriate answer and solution with the least
possible computational and time volume.
Therefore, contributions of this research are as follows: the application of Inclined
Planes system Optimization (IPO) algorithm with its modified and new version
(called MIPO) for the first time in this regard to intelligently estimate the covariance noise matrices of an EKF from an INS/GNSS system and achieve an optimal
navigation algorithm and desired results. In such a way that, over time and based on
the measured values reached to the filter, the correlation matrices of process noise
and the measurement noise, denoted by Q and R, respectively, are adapted to obtain
the least estimation error.
In simulations, based on technical and theoretical considerations governing an
assumed INS/GNSS navigation problem with two sets of IMU-based INSs with
unique features [18], IPO algorithms for estimating matrices Q and R are used. The
results of the algorithms are compared with two well-known algorithms, genetic algorithm (GA) and particle swarm optimization (PSO). The rest of the research is organized as follows: The navigation problem is presented in Sect. 2. Section 3 describes
the concept of intelligent optimization and the proposed algorithms. Section 4
presents the proposed approach with its considerations. The results and analysis
are reported in Sect. 5, and, finally, the work concluded in Sect. 6.
2 Expressing the Navigation Problem
Navigation involves moving and finding the way from one place to another. To
achieve this goal, a variety of equipment is available. Navigation covers a wide range
of applications, from industrial and commercial applications to military applications.
Metaheuristic Algorithms for Integrated Navigation Systems
51
Fig. 1 Block diagram of
guidance, navigation, and
control of a device [19]
Figure 1 shows a block diagram of the guidance, navigation, and control of a device
[19].
The four types of sample coordinate devices used in navigation systems are: inertia
device, ground device, navigation device, and body device. These coordinate devices
are used because INS mechanized outputs, including position, speed, and status, need
to be converted into user-understandable navigation information [20].
2.1 Inertial Navigation
The heart of an INS is its navigation processor, which uses IMU measurements using
mechanization. Mechanization refers to generating navigation responses from a set of
raw measurements obtained from sensors. This process begins with the initialization
and alignment of the system, followed by differential equations to provide navigation
responses. Figure 2 shows the mechanization process in general [20].
52
A. Mohammadi et al.
Fig. 2 The general process
of inertial navigation
mechanization [20]
2.2 Integrated Navigation
The final goal of integrated navigation is to estimate a state vector Y n of a moving
device in the current time step n by having a set of measurements (observations)
collected in time steps of 0, 1, …, N (Z n = {z0 , …, zn }). The state vector of a moving
device is in the form of Eq. (1):
T
Yn = n , n , Hn , VnN , VnE , VnD , n , n , n
(1)
where n , n , Hn , VnN , VnE , VnD , n , n , n are the latitude, longitude, altitude,
velocity in a northerly direction, velocity in an easterly direction, velocity in a downward vertical direction, yaw angle, roll angle, and pitch angle of the moving device,
respectively.
The state transition model (motion model) of the system is described as Eq. (2):
Yn = F(Yn−1 , Un−1 , Wn−1 )
(2)
where U n is the control input which is read values of IMU, W n is the process noise
which is independent of the past and present states of the system and is considered
due to the uncertainty in the movement of the moving device and the read values of
IMU.
The state measurement model is also:
Z n = H (Yn , Vn )
(3)
Metaheuristic Algorithms for Integrated Navigation Systems
53
where V n is a measurement noise that is independent of the past and current states
of the system as well as independent of process noise and is considered due to the
uncertainty in the read values of GPS.
The functions F and H are inherently nonlinear in the state transition model and
the measurement model, and the process and measurement noises are essentially
non-Gaussian and non-white. Therefore, the main problem of integrated navigation
is the modeling of these functions and noises, which is also considered by researchers
in the field of intelligent optimization based on artificial intelligence.
3 Optimization by Using Metaheuristic Algorithms
Optimization is the process by which optimal output/solution/result (maximum or
minimum) is made by setting the inputs of a problem or the specifications of a
component. Optimization in mathematics means resulting in the desired response,
which can be the minimum or maximum value of an index in the form of one or more
objective functions. In practice, the application of definitive solution approaches
and methods in optimization is not simply expressed; however, this has been met
by using a variety of stochastic optimization methods, also called heuristics and
metaheuristics.
A set of intelligent methods that complement each other to create robust and
inexpensive systems are categorized into soft computing techniques. Soft computing
includes methods such as neural networks, fuzzy logic, evolutionary computations
(including genetic algorithms), swarm intelligence, and heuristic and metaheuristic
approaches with random and probability-based reasoning, which can deal with uncertainty, ambiguity, incomplete or partial truth, machine learning, and optimization
problems. These features allow the creation of inexpensive intelligent systems with
a high degree of machine intelligence.
Many optimization problems in engineering are naturally more complex and challenging than can be solved by conventional optimization methods such as mathematical programming methods and the like. On the other hand, classical mathematics
methods have two basic forms: (a) they consider the local optimal point as the global
optimal point, and (b) each of these methods is used only for a specific problem.
Therefore, the main purpose of intelligent methods is to find the optimal answer to
engineering problems.
The use of metaheuristic algorithms for intelligent optimization leads to significant
improvements in reducing the time and computational volume required to solve the
desired problems. These algorithms are heuristic search methods mainly based on
counting methods, with the difference that they use additional information to guide
the search. These methods are quite general in terms of application and can solve
very complex problems. They mimic and exploit biological and physical processes.
Some of the optimization algorithms include genetic algorithm (GA) [21, 22], ant
colony optimization (ACO) [23], particle swarm optimization (PSO) [24], simulated
54
A. Mohammadi et al.
annealing (SA) [25], gravitational search algorithm (GSA) [26], differential evolution
(DE) [27–29], and gray wolf optimization (GWO) [30].
In the following, first, the two well-known competing algorithms of GA and PSO
are briefly described. Then, a detailed overview of the candidate algorithms “inclined
planes system optimization (IPO) algorithm [31]” and “its improved version MIPO
[32]” will be provided, respectively.
3.1 Genetic Algorithm
Genetic algorithm was first proposed by John Holland in 1975 [21]. This algorithm is one of the metaheuristic search algorithms. The genetic algorithms simulate
biological evolution systems.
In summary, GA begins by creating an initial response set as a random initial
population. Each member of the population is called a “chromosome”. Each chromosome contains a string of numbers called “genes”. The population at a given time
from the execution of the algorithm is also called “generation”. During the generation
process, genetic operators such as selection, crossover, and mutation are applied to
chromosomes with a probability, and a new generation of chromosomes is formed
called “the offspring”. Then the fitness of each offspring is calculated and by one of
the selection methods, better chromosomes are selected and passed on to the next
generation [21, 22].
In the execution of genetic algorithm (according to flowchart Fig. 3), the chromosomes of each generation try to better explore the search space of the problem by
using the available operators and increase their fitness, thus finding the best solutions
[21, 22].
3.2 Particle Swarm Optimization
Particle swarm optimization was first proposed by James Kennedy and Russell C.
Eberhart in 1995 [24]. This method is one of the powerful types of metaheuristic
algorithms adapted from the collective behavior of groups of animals such as birds
and fish.
According to the flowchart in Fig. 4, the algorithm begins with the initialization
of a random population. In PSO, each solution agent is called a “particle”, and each
of those particles is assigned a “velocity”. Each particle improves its position in the
problem-solving search space according to the two values of “Best Local Fitness
(Pbest )” and “Best Global Fitness (Gbest )”. The two main characteristics of particles, namely their velocity and position, are calculated by the following equations,
respectively:
vid (t + 1) = ωvid (t) + c1r1 Pibest (t) − pid (t) + c2 r2 G ibest (t) − pid (t)
(4)
Metaheuristic Algorithms for Integrated Navigation Systems
55
Fig. 3 Genetic algorithm
execution flowchart
pid (t + 1) = pid (t) + vid (t + 1)
(5)
where r 1 and r 2 are two random numbers. Parameters c1 and c2 are acceleration
constants to control local (personal) or global (social) searches. The inertia coefficient
ω is also another tool for controlling and creating the desired trade-off between exploration and exploitation processes of the algorithm, which is often reduced linearly
from a maximum value.
3.3 Inclined Planes System Optimization
The basis of this algorithm has been inspired by the dynamics of sliding motion on
a frictionless inclined plane. In IPO, a collection of agents (tiny balls) cooperate and
move toward better positions in the search space according to Newton’s second law
and equations of motion.
Consider a system with R balls (see Fig. 5). The position of the ball i is defined
by Eq. (6):
b = (bi1 , bi2 , . . . , bid , . . . , bir ),
f or i = 1, 2, . . . , R
(6)
56
Fig. 4 Workflow of PSO
algorithm
Fig. 5 A sample search
space with three balls in IPO
[31]
A. Mohammadi et al.
Metaheuristic Algorithms for Integrated Navigation Systems
57
That
≤ b j ≤ bmax
, 1≤ j ≤r
bmin
j
j
(7)
So that bid is the position of the ball i in the dimension d, in the r-dimensional
space. At a specified time, such as t, the angle between the ball i and j in dimension
d, ϕidj , is calculated as Eq. (8):
ϕidj (t)
= tan
−1
h j (t) − h i (t)
bid (t) − bdj (t)
,
f or d = 1, 2, . . . , r and i, j = 1, 2, . . . , R(i = j)
(8)
In Eq. (8), h i (t) is the value of the objective function (height) for the i-th ball at
time t. The amount and direction of acceleration for the ball i in time (iteration) t, in
dimension d, is determined by Eq. (9):
aid (t) =
R
U h j (t) − h i (t) sin(ϕidj (t))
(9)
j=1
where U(.) is the unit step function. Finally, Eq. (10) is used to update the position
of the balls:
bid (t + 1) = K 1 R AN D1 aid (t) t 2 + K 2 R AN D2 vid (t) t + bid (t)
(10)
where RAND1 and RAND2 are two random weights distributed uniformly over the
interval [0, 1]. Also, vi d (t) is the velocity of the ball i in dimension d, at time t, which
is calculated according to Eq. (11):
vid (t) =
d
bbest
(t) − bid (t)
t
(11)
d
where bbest
(t) is the ball with the lowest height (fitness) in the total iteration until the
current iteration. To control the algorithm’s search process, two important parameters
of K 1 and K 2 are defined in Eqs. (12) and (13). These two create a trade-off between
the two concepts of exploitation and exploration of the algorithm.
K 1 (t) =
K 2 (t) =
C1
1 + exp((t − S H I F T1 )SC AL E 1 )
(12)
C2
1 + exp(−(t − S H I F T2 )S H I F T2 )
(13)
58
A. Mohammadi et al.
Fig. 6 Execution process of
IPO algorithm
where C 1 , C 2 , SHIFT 1 , SHIFT 2 , SCALE 1 , and SCALE 2 are constants that are empirically determined for each function [31]. Figure 6 shows the execution process of
the IPO algorithm.
3.4 Modified Inclined Planes System Optimization
To reduce the complexity of the standard IPO, a modified version called MIPO
was proposed by Mohammadi et al. in 2017 [32]. Control parameters of MIPO are
two parameters of K 1 and K 2 , which change under the proposed damping/friction
coefficients K 1damp and K 2damp with iterations, as follows:
K 1 (t) = K 1damp
K 2 (t) = K 2damp
Max I ter − t
Max I ter
(14)
t
Max I ter
(15)
where MaxIter is the total number of iterations and t is the current iteration.
Metaheuristic Algorithms for Integrated Navigation Systems
59
So that high values of K 1 and lower values of K 2 , cause greater accelerations. This
leads to the greater motility of agents/balls. It means that, global searching or exploitation occurs with larger values of K 1 and lower measures of K 2 . On the contrary, if
the values of K 1 and K 2 become smaller and larger, respectively, exploration is
intensified.
In examining the structure of all the proposed algorithms, their simplicity and
relatively low relationships, along with their implementation and ease of use, increase
the motivation to apply them for the optimal design of INSs.
4 Metaheuristic Algorithms for Designing Integrated
Navigation Systems
The INS/GNSS integration problem is nonlinear. We look for an optimal nonlinear
filter in the form of intelligently estimating the values of the process and measurement noise covariance matrices (Q and R) for accurately determining the navigation response. Therefore, the proposed solution is to improve the efficiency of the
Kalman integration filter by the algorithms IPO and MIPO for designing an intelligent
navigation system.
The optimization is done in two ways: one is the optimal estimation of Kalman
parameters, and the other is the definition of an intelligent mechanism in the objective function of the problem. The proposed approach improves the performance
of the INS/GNSS system only by adjusting the noise covariance matrices in the
integrated navigation algorithm without changing the structure of the assumed integration problem simulated in [18]. Therefore, the basis of comparison is based on
the results extracted from the reference navigation system in [18]. To confirm the
performance and outputs of the proposed approaches, the results are compared with
two well-known rival algorithms, GA and PSO, and analyzes are reported in detail.
The implementation considerations of the assumed INS/GNSS problem are
entirely consistent with [18]. In [18], a simulation framework for low-cost INSs,
abbreviated NaveGo, is embedded as a MATLAB toolbox. All data and route specifications, inertia sensors, GPS, navigation and integration mechanism, calculations,
problem-solving relations, etc., are also described in detail in [18] and included in its
MATLAB source. In the proposed approach, instead of replacing the values obtained
from the GPS error profile in the main diameter of the measurement matrix R and
instead of inserting the noise and bias values of each sensor in the process covariance matrix Q, the values estimated by the proposed metaheuristic methods are used.
Given that in [18], the root-mean-square error (RMSE) for both actual and simulated
INS systems per MEMS IMU module are compared and reported, the sum of these
two errors as a weighted sum function [with the same weight and simple algebraic
sum as Eq. (16)] is minimized as a single-objective function of the problem.
Objective Function (O F) = sum([RMSE(IMU1), RMSE(IMU2)])
(16)
60
A. Mohammadi et al.
Fig. 7 Intelligent
optimization of the
INS/GNSS integrated
navigation problem using
IPO algorithms
where RMSE is the difference between the simulated INS/GNSS system and the
actual registered reference values as well as the independent GNSS system with the
reference for the two sets of IMU-based on MEMS technology in [18]. RMSE values
include the difference of all values of performance parameters, including latitude and
longitude, altitude, roll, pitch, yaw, and velocity in three directions. The flowchart
of Fig. 7 shows the general framework of the proposed approach.
The purpose of optimizing matrices (Q) and (R) is to estimate their elements so
that the fitness error of Eq. (17) is minimized, so the search agents’ vector is defined
as follows:
X T = x1R , x2R , . . . , x6R , x7Q , . . . , x NQ
(17)
The solution is an N-dimensional vector. N is the number of search agents and
is equal to the sum of the main diameters, two measurement and process noise
covariance matrices, and is considered in the proposed optimization algorithms.
Metaheuristic Algorithms for Integrated Navigation Systems
61
5 Results
In this section, outputs are reported according to the following indicators and criteria:
trajectory (in terms of latitude, longitude, and altitude), attitude (roll, pitch, and yaw),
errors of roll, pitch and yaw, changes in velocity (received from GPS including
components north velocity V N , east velocity V E , and vertical velocity V D ), velocity
errors, variations in latitude, longitude, and altitude, and finally errors of latitude,
longitude, and altitude. All of these are extracted by simulating the movement of
a vehicle on a specific trajectory [18], and in three formats (REF or True (correct
vehicle route), single GPS output, and using two modules IMU1 and IMU2) are
displayed distinctly and symmetrically.
Here, the INS/GNSS navigation is formulated as a single-objective optimization
problem and is simulated for two proposed IPO and MIPO algorithms along with
two popular competitors GA and PSO. Due to the very high volume of data collected
in the reference problem [18], implementations have led to a lot of time spent.
Therefore, several computer systems have been used. Simulations have been carried
out in MATLAB (versions R2015b and R2019b) on a laptop and three other computer
systems with the following specifications:
• Intel (R) Core (TM) i3-2348M CPU@2.30 GHz, 6 GB RAM under Windows 7
Ultimate
• Intel (R) Core (TM) i5-6500 CPU@3.20 GHz, 8 GB RAM under Windows 10
Pro
• Intel (R) Core (TM) i7-4790 CPU@3.60 GHz, 16 GB RAM under Windows 10
Pro
• Intel (R) Core (TM) i5-3570 CPU@3.40 GHz, 20 GB RAM under Windows 10
Pro.
Each algorithm is executed for five independent trials/runs with a different number
of iterations. Graphic results are presented for the best trial of each algorithm and
numerical and statistical analyzes for all of them. Iterations are selected to estimate
the operational success of each method for low to high iterations. Control parameters
of the algorithms are listed in Table 2. Control values have been adopted based on
numerous experimental results and similar studies. In PSO, the inertia coefficient w
decreases linearly with a friction factor of wdamp corresponding to the iteration step of
the algorithm. In Table 2, some values are set in an interval (such as [0.9–1]) to obtain
the desired response from the algorithm. In contrast, some algorithms achieved the
desired results for fixed control values for all trials. This stable and successful performance for minimal changes (or fixed) in control parameters indicates the superiority
and robustness of the method in the optimized problem.
The range of variations of the design variables in the form of vectors of minimum
and maximum selectable intervals (equal to 18 variables including the main diameters
of the matrices R and Q, respectively) in Eqs. (18) and (19) is considered in a specific
and limited way:
62
A. Mohammadi et al.
Table 2 Control parameters for algorithms
Parameter
Algorithm
GA
Iterations
PSO
IPO
MIPO
100, 200, 300, 400, 500
Population
50
Dimensions
18
Crossover (rate)
One-point (0.95)
–
–
–
Mutation (rate)
Uniform (0.05)
–
–
–
Selection
Roulette wheel
–
–
–
c1 , c2
–
2
–
–
ω
–
0.99
–
–
C1
–
–
0.2
–
C2
–
–
1.7
–
SHIFT 1
–
–
1
–
SHIFT 2
–
–
80
–
SCALE 1
–
–
0.2
–
SCALE 2
–
–
0.73
–
K 1damp
–
–
–
0.005–1
K 2damp
–
–
–
0.9–2
X Min = [0, 0, 0, 0, 0, 0, 1−10 , 1−10 , 1−10 , 1−10 , 1−10 , 1−10 ,
1−10 , 1−10 , 1−10 , 1−10 , 1−10 , 1−10 ]
X Max = [1, 1, 1, 100, 100, 100, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
(18)
(19)
In the results, the best numerical values are displayed in bold. The Outputs are
reported in the form of estimated values of design variables [X T of Eq. (17)], convergence curves, statistical analysis of fitness values of the objective function, execution
times, numerical values of performance criteria of the problem, all based on 5 trials of
all algorithms, and graphical results for the best trial of each algorithm, respectively.
Due to the long-running time, it is reported by the hour. The values reported for the
performance criteria of the problem represent the output accuracy of the INS/GNSS
integrated system optimized by the proposed intelligent algorithms. So that the value
in each row corresponding to the column of an algorithm, indicates the system error
in terms of the parameters of attitude (roll, pitch, and yaw), velocities, and position
(longitude, longitude, and latitude).
The convergence curves of the algorithms for 5 trials are also shown in Fig. 8.
Figure 9 shows the estimated design variables per 5 runs for all algorithms.
In Table 3, general evaluations of the results for all trials with iterations of 100,
200, 300, 400, and 500 together in the form of using statistical indicators minimum
Metaheuristic Algorithms for Integrated Navigation Systems
(a)
(b)
(c)
Fig. 8 Convergence curves of algorithms for iterations. a 100, b 200, c 300, d 400, and e 500
63
64
A. Mohammadi et al.
(d)
(e)
Fig. 8 (continued)
(best), maximum (worst), average (mean), and variance are presented. Figure 10
shows the best routing (trajectory) of each algorithm during the 5 independent runs.
Finally, in Table 4, to evaluate the overall performance of the algorithms, a final
ranking for each algorithm is presented. Table 4 shows the numerical superiority of
the statistical indicators of the algorithms in Table 3 based on each of the performance
criteria and dividing by the total number of them. This analysis represents the overall
performance of each algorithm in terms of a combination of statistical indicators
and performance metrics that confirm the success and superiority of algorithms,
respectively.
In Table 4, the results of the total rating calculations of each algorithm based
on the use of each of the inertial navigation systems IMU1 and IMU2 along with
the ratings obtained from the statistical results of runtimes and fitness values of the
objective function and finally an overall and final ranking are reported in the last row.
As can be seen in Table 4, the final ranking indicates that the success and operational
Metaheuristic Algorithms for Integrated Navigation Systems
65
Fig. 9 Estimated design variables for all algorithms
superiority of the algorithms in the given integrated navigation problem has been
achieved through the use of IPO, PSO, GA, and finally, MIPO, respectively.
6 Conclusions
In this research, the intelligent optimization of INS/GNSS integrated navigation
systems was investigated and analyzed by using some metaheuristic optimization
algorithms. The implementations were performed under MATLAB, and the resulting
outputs were presented in detail. Therefore, for the first time, the capabilities and
robustness of the IPO algorithm were used along with its improved version called
MIPO.
In addition to comparing the outputs to the reference system, the results were
evaluated using two well-known and common evolutionary algorithms, GA and PSO.
The performance of the proposed methods was measured to intelligently estimate the
numerical values of process and measurement noise covariance matrices R and Q.
66
A. Mohammadi et al.
Table 3 Statistical evaluation of algorithms for five independent trials
Algorithms →
GA
PSO
IPO
MIPO
9.6630
7.6625
6.5175
7.3427
Max
47.7810
49.9076
12.1359
53.3394
Mean
23.3065
17.1783
8.7841
32.1500
Var
293.64
335.83
6.182
488.36
Min
13.0122
11.1617
12.9660
13.1009
Max
83.5560
81.173
71.7840
73.4099
Mean
39.7600
39.3454
37.3376
38.7888
Var
801.01
735.8
711.25
666.04
Min
1.6620
1.5881
1.5832
6.9383
Max
26.3160
37.8670
8.7665
34.3600
Mean
10.1516
13.9757
5.4821
13.6410
Var
102.77
207.62
8.99
135.13
Min
4.0077
1.9841
2.0835
4.9987
Max
39.9580
55.8410
8.5539
51.4550
Mean
13.3993
18.2680
4.44635
16.3782
Var
227.56
461.4
6.1403
389.33
Min
21.4530
50.2770
5.8749
43.0950
Max
58.5830
67.1810
58.3660
58.8660
Mean
45.4306
58.5870
32.8872
49.8378
Var
214.43
38.16
662.74
42.993
Min
0.0468
0.0479
0.0472
0.0480
Performance results ↓
Fitness
Min
Runtime (h)
For using IMU1 (ADIS16405)
Roll°
INS/GNSS
Pitch°
Yaw°
Velocity
(m/s)
N
GNSS
INS/GNSS
E
GNSS
INS/GNSS
Max
0.0491
0.0492
0.0494
0.0491
Mean
0.0481
0.0485
0.0481
0.0483
Var
8.4e−07
2.4e−07
7.3e−07
2.03e−07
Min
0.0628
0.0622
0.0559
0.0751
Max
0.2527
0.3129
0.0816
0.2882
Mean
0.1182
0.1510
0.0665
0.1631
Var
0.0067
0.0131
0.0001
0.0069
Min
0.0481
0.0480
0.0479
0.0477
Max
0.0498
0.0486
0.0494
0.0487
Mean
0.0486
0.0482
0.0484
0.0482
Var
4.8e−07
0.6e−07
3.9e−07
1.9e−07
Min
0.0601
0.0590
0.0624
0.0739
Max
0.2153
0.3234
0.0841
0.3132
(continued)
Metaheuristic Algorithms for Integrated Navigation Systems
67
Table 3 (continued)
D
GNSS
INS/GNSS
Latitude (m)
GNSS
INS/GNSS
Longitude (m)
GNSS
INS/GNSS
Altitude (m)
GNSS
INS/GNSS
Mean
0.1216
0.1430
0.0740
Var
0.0048
0.0122
0.0001
0.1397
0.0098
Min
0.0476
0.0474
0.0467
0.0470
Max
0.0488
0.0488
0.0491
0.0487
Mean
0.0479
0.0481
0.0482
0.0481
Var
2.4e−07
2.6e−07
8.7e−07
6.4e−07
Min
0.0417
0.0419
0.0353
0.0613
Max
0.2697
0.3818
0.1049
0.3531
Mean
0.0999
0.1258
0.0543
0.1455
Var
0.0096
0.0213
0.0008
0.0140
Min
4.512
4.5986
4.5603
4.6306
Max
4.7393
4.7258
4.7599
4.7337
Mean
4.6696
4.6504
4.6526
4.6814
Var
0.0047
0.0035
0.0065
0.0021
Min
0.5376
0.4541
0.4545
0.5016
Max
0.7827
0.7077
0.6917
0.6980
Mean
0.6341
0.5709
0.6235
0.5868
Var
0.0104
0.0095
0.0100
0.0085
Min
4.6119
4.6545
4.6195
4.5769
Max
4.8234
4.8089
4.7152
4.7662
Mean
4.6981
4.7202
4.6751
4.6918
Var
0.0063
0.0034
0.0015
0.0052
Min
0.4823
0.4649
0.4214
0.5223
Max
1.0322
0.6117
0.7482
0.6173
Mean
0.6509
0.5514
0.5415
0.5770
Var
0.0521
0.0040
0.0155
0.0015
Min
9.0730
9.1733
9.1892
9.3156
Max
9.4774
9.5190
9.5028
9.7686
Mean
9.3323
9.3787
9.3690
9.4442
Var
0.0247
0.0221
0.0175
0.0349
Min
0.8217
0.7811
0.7925
0.8545
Max
1.1181
1.1967
1.6459
1.2862
Mean
0.9459
1.0044
1.0445
1.0433
Var
0.0150
0.0308
0.1187
0.0273
Min
1.4017
1.4178
1.2158
1.1673
Max
9.3368
10.0740
2.7495
11.4730
For using IMU2 (ADIS16488)
Roll°
INS/GNSS
(continued)
68
A. Mohammadi et al.
Table 3 (continued)
Pitch°
Yaw°
Velocity
(m/s)
N
GNSS
INS/GNSS
E
GNSS
INS/GNSS
D
GNSS
INS/GNSS
Latitude (m)
GNSS
Mean
4.1088
4.0736
1.6393
6.0718
Var
14.181
15.06
0.3937
21.246
Min
1.4671
1.3145
1.4133
1.3850
Max
7.4989
8.1514
7.7034
9.4343
Mean
3.8158
3.6761
2.8484
5.3832
Var
8.841
9.073
7.4004
13.495
Min
7.2064
7.1950
2.8955
3.1076
Max
55.6320
52.9750
50.1520
58.4460
Mean
26.4415
23.5994
16.1993
36.3517
Var
472.95
372.212
369.543
717.177
Min
0.0478
0.0479
0.0472
0.0480
Max
0.0491
0.0492
0.0494
0.0491
Mean
0.0481
0.0485
0.0481
0.0483
Var
8.4e−07
2.4e−07
7.3e−07
2.03e−07
Min
0.0528
0.0558
0.0541
0.0594
Max
0.0879
0.0862
0.0636
0.0793
Mean
0.0656
0.0685
0.0579
0.0701
Var
2.1e−04
2.04e−04
0.14e−04
0.95e−04
Min
0.0481
0.0480
0.0479
0.0477
Max
0.0498
0.0486
0.0494
0.0487
Mean
0.0486
0.0482
0.0484
0.0482
Var
4.8e−07
0.6e−07
3.9e−07
1.9e−07
Min
0.0570
0.0588
0.0539
0.0530
Max
0.1502
0.1008
0.0669
0.0952
Mean
0.0837
0.0712
0.0604
0.0749
Var
0.0016
0.0003
0.00001
0.0004
Min
0.0476
0.0474
0.0467
0.0470
Max
0.0488
0.0488
0.0491
0.0487
Mean
0.0481
0.0481
0.0482
0.0481
Var
3.4e−07
2.6e−07
8.7e−07
6.4e−07
Min
0.0381
0.0382
0.0357
0.0397
Max
0.0619
0.0491
0.1084
0.0587
Mean
0.0445
0.0431
0.0514
0.0483
Var
0.97e−04
0.17e−04
8.96e−04
0.66e−04
Min
4.5612
4.5986
4.5603
4.6306
Max
4.7393
4.7258
4.7599
4.7337
Mean
4.6696
4.6504
4.6526
4.6814
(continued)
Metaheuristic Algorithms for Integrated Navigation Systems
69
Table 3 (continued)
INS/GNSS
Longitude (m)
GNSS
INS/GNSS
Altitude (m)
GNSS
INS/GNSS
Var
0.0047
0.0035
0.0065
0.0021
Min
0.5389
0.4244
0.4552
0.5011
Max
0.7859
0.7062
0.6926
0.6995
Mean
0.6366
0.5646
0.6238
0.5889
Var
0.0108
0.0113
0.0099
0.0099
Min
4.6119
4.6545
4.6195
4.5769
Max
4.8234
4.8089
4.7152
4.7662
Mean
4.6981
4.7202
4.6751
4.6918
Var
0.0063
0.0034
0.0015
0.0052
Min
0.4822
0.4600
0.4198
0.5263
Max
1.0306
0.6160
0.7466
0.6206
Mean
0.6486
0.5499
0.5411
0.5797
Var
0.0523
0.0042
0.0154
0.0015
Min
9.0730
9.1733
9.1892
9.3156
Max
9.4774
9.5190
9.5028
9.7686
Mean
9.3323
9.3787
9.3690
9.4442
Var
0.0247
0.0221
0.0175
0.0349
Min
0.8218
0.7808
0.7945
0.8539
Max
1.1170
1.2258
1.6460
1.2863
Mean
0.9388
1.1012
1.0449
1.0388
Var
0.0150
0.0341
0.1185
0.0285
From the general assessment, it can be acknowledged that the use of metaheuristic
IPO approaches to optimally design integrated navigation systems is successful and
can be a good candidate compared to the computational volume of other mathematical
and theoretical methods.
Some suggestions for future work include: (1) replacing navigation algorithms
with techniques based on reinforcement or deep learning; (2) utilizing multiobjective optimization versions of evolutionary and metaheuristic algorithms; (3)
using the potential and capabilities of other intelligent algorithms; (4) applying hybrid
approaches such as fuzzy logic and artificial neural networks to improve performance
criteria of navigation problems.
70
Fig. 10 Trajectories of
optimized navigation
systems by the best trial of a
GA, b PSO, c IPO, and d
MIPO algorithms
A. Mohammadi et al.
Metaheuristic Algorithms for Integrated Navigation Systems
71
Table 4 Final ranking of algorithms (based on Table 3)
Algorithms →
GA
PSO
IPO
MIPO
Fitness
12/4
11/4
4/4
16/4
Runtime
23/4
16/4
14/4
21/4
For IMU1
171/60
202/60
149/60
223/60
For IMU2
233/60
196/60
165/60
236/60
Total
439/128
425/128
332/128
492/128
Ranking ↓
The bold values represent the best outputs
References
1. G. Minkler, J. Minkler, Theory and Applications of Kalman filtering (Magellan Book Company,
1993)
2. V. Sathiya, M. Chinnadurai, Evolutionary algorithms-based multi-objective optimal mobile
robot trajectory planning. Robotica 37, 1363–1382 (2019)
3. S.M.J. Jalali, A. Khosravi, P.M. Kebria, R. Hedjam, S. Nahavandi, Autonomous robot navigation system using the evolutionary multi-verse optimizer algorithm, in 2019 IEEE International
Conference on Systems, Man and Cybernetics (SMC) (IEEE, 2019), pp. 1221–1226
4. Q. Liu, Y. Li, L. Liu, A 3D simulation environment and navigation approach for robot navigation via deep reinforcement learning in dense pedestrian environment, in 2020 IEEE 16th
International Conference on Automation Science and Engineering (CASE) (IEEE, 2020),
pp. 1514–1519
5. L. Cong, S. Yue, H. Qin, B. Li, J. Yao, Implementation of a MEMS-Based GNSS/INS integrated
scheme using supported vector machine for land vehicle navigation. IEEE Sens. J. 20, 14423–
14435 (2020)
6. J.-H. Yi, M. Lu, X.-J. Zhao, Quantum inspired monarch butterfly optimisation for UCAV path
planning navigation problem. Int. J. Bio-Inspired Comput. 15, 75–89 (2020)
7. M.G. Bellemare et al., Autonomous navigation of stratospheric balloons using reinforcement
learning. Nature 588, 77–82 (2020)
8. N. Al Bitar, A.I. Gavrilov, Neural networks aided unscented Kalman filter for integrated
INS/GNSS systems, in 2020 27th Saint Petersburg International Conference on Integrated
Navigation Systems (ICINS) (IEEE, 2020), pp. 1–4
9. J. Wang, Z. Ma, X. Chen, Generalized dynamic fuzzy NN model based on multiple fading
factors SCKF and its application in integrated navigation. IEEE Sens. J. 21, 3680–3693 (2021)
10. F. Gul et al., Meta-heuristic approach for solving multi-objective path planning for autonomous
guided robot using PSO–GWO optimization algorithm with evolutionary programming. J.
Ambient Intell. Humaniz. Comput. 12, 7873–7890 (2021)
11. P. Zieliński, U. Markowska-Kaczmar, 3D robotic navigation using a vision-based deep
reinforcement learning model. Appl. Soft Comput. 110, 107602 (2021)
12. D. Gao, X. Lyu, F. Qin, L. Chang, B. Hu, A real time gravity compensation method for high
precision INS based on neural network, in 2021 28th Saint Petersburg International Conference
on Integrated Navigation Systems (ICINS) (IEEE, 2021), pp. 1–5
13. S. Wen, Z. Wen, D. Zhang, H. Zhang, T. Wang, A multi-robot path-planning algorithm for
autonomous navigation using meta-reinforcement learning based on transfer learning. Appl.
Soft Comput. 110, 107605 (2021)
14. N. Al Bitar, A. Gavrilov, A novel approach for aiding unscented Kalman filter for bridging
GNSS outages in integrated navigation systems. Navigation 68, 521–539 (2021)
72
A. Mohammadi et al.
15. F. Yan, S. Li, E. Zhang, J. Guo, Q. Chen, An adaptive nonlinear filter for integrated navigation
systems using deep neural networks. Neurocomputing 446, 130–144 (2021)
16. Y. Wu, A survey on population-based meta-heuristic algorithms for motion planning of aircraft.
Swarm Evol. Comput. 62, 100844 (2021)
17. E.P. Herrera, H. Kaufmann, Adaptive methods of Kalman filtering for personal positioning
systems, in Proceedings of the 23rd International Technical Meeting of the Satellite Division
of The Institute of Navigation (ION GNSS 2010) (2010), pp. 584–589
18. R. Gonzalez, J.I. Giribet, H.D. Patino, NaveGo: a simulation framework for low-cost integrated
navigation systems. J. Control Eng. Appl. Inform. 17, 110–120 (2015)
19. J.F.W. Georgy, Advanced nonlinear techniques for low cost land vehicle navigation. Queen’s
University (2010)
20. K. Abdul Rahim, Heading drift mitigation for low-cost inertial pedestrian navigation. University
of Nottingham (2012)
21. J.H. Holland, Adaptation in Natural and Artificial Systems (The University of Michigan Press,
1975)
22. D.E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning (AddisonWesley Publishing Company, 1989)
23. M. Dorigo, V. Maniezzo, A. Colorni, Ant system: optimization by a colony of cooperating
agents. IEEE Trans. Syst. Man Cybern. Part B 26, 29–41 (1996)
24. J. Kennedy, R. Eberhart, Particle swarm optimization, in Proceedings of ICNN’95-International
Conference on Neural Networks, vol. 4 (IEEE, 1995), pp. 1942–1948
25. S. Kirkpatrick, C.D. Gelatt, M.P. Vecchi, Optimization by simulated annealing. Science 220,
671–680 (1983)
26. E. Rashedi, H. Nezamabadi-Pour, S. Saryazdi, GSA: a gravitational search algorithm. Inf. Sci.
179, 2232–2248 (2009)
27. R. Storn, K. Price, Differential evolution—a simple and efficient heuristic for global
optimization over continuous spaces. J. Glob. Optim. 11, 341–359 (1997)
28. K. Price, R.M. Storn, J.A. Lampinen, Differential Evolution: A Practical Approach to Global
Optimization (Springer, Berlin, 2005)
29. S. Das, P.N. Suganthan, Differential evolution: a survey of the state-of-the-art. IEEE Trans.
Evol. Comput. 15, 4–31 (2011)
30. S. Mirjalili, S.M. Mirjalili, A. Lewis, Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61 (2014)
31. M.H. Mozaffari, H. Abdy, S.H. Zahiri, IPO: an inclined planes system optimization algorithm.
Comput. Inform. 35, 222–240 (2016)
32. A. Mohammadi, S.H. Zahiri, IIR model identification using a modified inclined planes system
optimization algorithm. Artif. Intell. Rev. 48, 237–259 (2017)
Security Threats in Flying Ad Hoc
Network (FANET)
Safia Lateef, Muhammad Rizwan, and Muhammad Abul Hassan
Abstract In the last few decades, the progress in use of technology in electronics and
avionics system mainly covering the cost reduction and miniaturization of devices
uplifted the UAV (Unmanned Ariel Vehicles). As an emerging technology having
lifesaving characteristics has inspired many industrialists including civil and military.
Hence, it has been a focus of different researchers for quite years. Thus, it opens up to
new era of network technology between Unmanned Ariel Vehicles (UAV) and ground
stations known as Flying Ad hoc Network (FANET). The basic working of Flying Ad
hoc network (FANET) is developed using the Mobile Ad hoc Networks (MANETs)
principle. In 2001, Aerial Network was introduced under the applications of “droneto-drone ad hoc mobile communication and networking”. This enabled the formation
of networks that could be relayed among Aerial vehicles. Even though flying IoT has
proved to be quite useful over the years however, there are some aspects to it that make
it quite vulnerable to network attacks. There may be different kinds of attackers with
different intentions. Moreover, there are countless types of attacks that can be made
which interfere with the internal system of Flying-things making the user vulnerable.
In this review paper, Major challenges faced by FANET with respect to its security
and related research that has been carried out to overcome these challenges in Flying
network is discussed. Finally, we have concluded by highlighting the importance of
secure systems especially FANET in order to provide security to its users.
Keywords FANET · UAV · Ad hoc network · Security · Attacks · Threats ·
Solutions
S. Lateef (B) · M. Rizwan
Department of Computer Science, Kinnaird College for Women, Lahore, Pakistan
M. Rizwan
e-mail: muhammad.rizwan@kinnaird.edu.pk
M. A. Hassan
Department of Computer and Technology, Abasyn University, Peshawar, Pakistan
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_5
73
74
S. Lateef et al.
1 Introduction
Flying Ad hoc Network helps in the reduction of autonomous accidents by improving
traffic flows in Intelligent Transportation system (ITS). This is achieved by instilling
the air vehicular with the required information or simply by providing the information to the pilots/controllers. However, sometimes Flying-IoT may become as
deadly as lifesaving it is originally meant to be. Any nonessential changes in the
real-time information provided to the aerial vehicle may prove to be fatal to the
pilots/controllers in ITS. Therefore, to ensure secure information and smooth functioning of the system, all the security threats must be eliminated completely from the
Unmanned Aerial System (UAS) [1]. Nowadays, pilots/controllers are highly vulnerable to accidents. They could either be too engrossed in their mobile phones or lose
control due to mechanical, electronic and electrical faults, whatever the reason, it was
vital to introduce such technology that would minimize the number of accidents. So,
in order to provide increased safety, researchers and experts developed a wireless
system. This wireless network introduces the option of wireless environment to its
users so to allow them to fly freely, leading to dynamic topologies. In this idea, the
aerial vehicles become a part of a particular area in a network of communication. To
solve this problem, the mechanism of MANET (Mobile Ad hoc Network) is used by
many researchers which is a wireless ad hoc network which consists of independent
and wireless devices that move about in random directions. FANET was discovered
within the field of MANET and uses similar methodologies in particular of MANET
and VANET. Figure 1 shows the hierarchy of MANET, VANET and FANET. The
movement of each node in flying aerial network is random to directions which as a
result frequently changes the links amongst drones. The data transmission is done in
such a way that nodes communicate with other existing nodes, which in turn communicate with the ground station. In order to maintain the flow of data, the devices on
Fig. 1 Hierarchy of MANET, VANET and FANET
Security Threats in Flying Ad Hoc Network (FANET)
75
the nodes must always be under working conditions and must send data continuously [2]. The users of Flying-IoT gains profits from several applications that are
significantly classified into active flying safety, traffic efficiencies, speed control and
cooperative navigation in ITS [3]. In FANET however, it is vital to ensure the safety of
Unmanned Aircraft Systems (UAS) against malicious attacks. In early days of Aerial
Network, several researchers performed detailed studies on the wireless system and
had explored the security threats that FANET is prone to. Other researchers focused
on defining the security infrastructures and formalize protocols. But to this date,
the security threats in UAS are yet to be fully explored [4]. The core focus of this
paper lies on the security challenges and threats faced by FANET and the possible
solutions provided by different papers are portrayed in the following sections. The
main objective of this paper is to provide a collective analysis of several research
and review based studies and explain the core concepts of each one of them.
2 Overview of FANET
The Flying Ad hoc Network (FANET) has emerged as an alternative access technology for regions that have no fixed infrastructure or are very hard to reach. Flying
Ad hoc network (FANET) is a self-organizing wireless network that enables inexpensive, flexible, and easy-to-deploy flying nodes, such as UAVs, to communicate among
themselves in the absence of fixed network infrastructure on the ground as shown
in Fig. 2. Aerial vehicles contains sensors, GPS, processor, camera, transmitters and
antennas.
UAV perform two tasks i.e. establishes U2U link with other UAVs and collect the
information and data then relays the same information to ground control station via
U2I link. Following are the types of communication used in FANET:
1.
2.
3.
4.
Inter-plane Communication: Two planes Communicate with each other.
Intra-plane Communication: Occur between the UAVs of same plane.
Ground Station Communication: Occur between UAV and ground station.
Ground Sensor Communication: Occur between the UAVs of any plane with
the sensors installed on ground.
FANETs are generally used to provide connectivity to hard-to-reach places in
regions where there have been natural disasters or even for military applications.
After a catastrophic event (such as an earthquake, hurricane, tsunami, and dam
breach), traditional network infrastructures can suffer damage and be subject to automatic shutdowns. However, through a FANET configuration, they could be employed
to restore and provide sufficient connection and communication to the network in
isolated areas.
In FANETS, one of the UAV is directly linked to the infrastructure, whereas the
rest of the UAVs in the network work on multi-hop communication in which each
node act as a hop count or relay. Following are some of the characteristics of FANET
and are also shown in Table 1.
76
S. Lateef et al.
Fig. 2 Architecture of FANET
Table 1 Characteristic of
MANET, VANET FANET
Characteristic
MANET
VANET
FANET
Node mobility
Low
High
Very high
Mobility model
Random
Regular
Random
Node density
Small
Large
Very small
Network topology
Low
High
Very high
Power consumption
Required
Not required
Required
2.1 Network Topology
In FANET, the nodes have high mobility which changes the network topology again
and again [5, 6].
2.2 Mobility Models
The mobility model for some flights is pre-planned and the map is relocated at every
change. Some of the model uses random speed and direction [6].
Security Threats in Flying Ad Hoc Network (FANET)
77
2.3 Node Mobility
The degree of mobility in FANET is more important than the other two ad hoc
networks as the nodes can change their location in shorter time interval which can
cause communication problem [7, 8].
2.4 Node Density
The node density in FANET is much lower than that of MANET and VANET because
the nodes should be scattered in the sky with large distances between them, according
to the nature of flying [9, 10].
2.5 Localization
Localization data of nodes in FANET should be updated for short intervals of time
due to high mobility degree. So, it makes easier to locate the UAV [11].
2.6 Power Consumption
Contrary to MANET, large UAV are not power sensitive used in FANET. However,
the mini-UAVs their electric capacity is limited [5].
2.7 Radio Propagation Model
In FANETS, UAVs uses Line of Sight (LoS) between them and the ground station
[12].
3 Literature Review
Over the years, many state-of-the-art publications have been made on FANET as
a whole. Several researchers studied the mechanism of FANET, others studied its
usefulness to the society. What gained most attention from researchers however is
the security threats to the entire system of FANET. Studies such as İlker Bekmezci,
Eren Şentürk, Tolgahan Türker in (SECURITY ISSUES IN FLYING AD HOC
78
S. Lateef et al.
NETWORKS) [13], Sumra, Irshad and Sellappan, P. and Abdullah, Azween and Ali,
Ahmad in (Security issues and Challenges in MANET-VANET-FANET: A Survey)
[14], Amira Chriki, Haifa Touati, Hichem Snoussi, Farouk Kamoun in (FANET:
Communication, Mobility models and Security issues) [15], Amartya Mukherjee,
Vaibhav Keshary, Karan Pandya, Nilanjan Dey, Suresh Chandra Satapathy in (Flying
Ad hoc Networks: A Comprehensive Survey [16], Amit Kumar Goyal, Gaurav
Agarwal, Arun Kumar Tripathi in (Network Architectures, Challenges, Security
Attacks, Research Domains and Research Methodologies in VANET: A Survey)
[17], Sparsh Sharma, Ajay Kaul in (VANETs Cloud: Architecture, Applications,
Challenges, and Issues) [18], and many more have focused their studies on the challenges and security threats in the wireless network system, FANET. The major attacks
on FANET can be categorized into 4 subcategories i.e. [1]. In addition to using a
small number of UAVs, UAV samples can work together to perform complex tasks
over very large areas, especially for monitoring and surveillance applications, but fly
in special networks. (FANET) When most UAVs communicate ad hoc, connectivity
and coverage can be achieved in situations of constraints of the terrestrial network:
remote sites, highly mobile and distributed sites. They can be expanded effectively.
However, while UAV interference must be effectively mitigated for successful UAVs
operation-built systems, UAV motion, supply supervision, and governor are primarily
based on UAV concentration. There is a problem with the variety of types. The same is
true for interoperability and positioning between different wireless networks. In addition, not only is the UAV’s capacity limited in terms of network load and onboard
processing, but the need for engine power and flying device is crucial real-world
aspects limiting large-scale use of UAV.
1.
2.
3.
4.
Threats to the wireless interface.
Threats to software and hardware.
Treats to sensor inputs in the aerial vehicle.
Threats behind the wireless access.
In the first category, the attacker tries in obtaining the pilots/controllers information by tracing him. This places the nodes at risk. Moreover, sometimes the attacker
tries to defer the signals that are supposed to reach the pilots/controllers. This may
be done by physically jamming the system. Another attack on the wireless interface
is the one in the attacker uses a trial-and-error technique to obtain personal information such as passwords or IDs of the pilots/controllers known ad brute force attack.
In the second category, the author mentions the possible ways in which attackers
can introduce threats into the hardware/software such as—Tampering Hardware,
Routing Attack, Timing attack, Replay attack and much more. The third category
attacks include Illusion attack and Jamming attack. Finally, threats behind wireless
access. Fourth may be—Unauthorized access, Session Hijacking or Repudiation. In
another study, Miscellaneous threats to the system to FANET are introduced in the
following categories—timing attack, home attack, traffic analysis, social attack, man
in the middle attack, bogus information etc. These attacks are directly made with
the intention of interfering with the confidentiality of the user. Confidentiality may
be accessed by various methods such as unlawful collection of messages through
Security Threats in Flying Ad Hoc Network (FANET)
79
overhearing or gathering the location information of the pilots/controllers using the
transmission of broadcast messages.
4 Security in FANET
Several studies have critically analyzed the challenges faced by FANET. The types
of attacks and different malicious attackers have been studied in great depth. FANET
is divided into three components i.e. [10]
1.
2.
3.
Mobile domain
Infrastructure domain
Generic domain.
Whilst performing an attack, all attackers target one or more of these domains as
their target location. It is therefore vital for the authorities that control the production
of FANET to introduce core changes into the major domain of the system. attackers
make use of any existing cracks of vulnerabilities in a system therefore, it is crucial
to reinforce the internal and external system of FANET in order to gain maximum
security. Security breaches are faced by all greats systems and organizations. One
such example is Windows XP that introduced a feature known as “Data Execution
Prevention” (DEP) to avoid security attacks such as buffer overflows. DEP generates
warnings to users if access is trying to be made through applications. So, in case of
attempts made by any source to carry out code from a page that is secured by DEP,
a violation of memory access exception takes place. If the violation is not answered
or handled, the calling procedure is called off. Therefore, it is absolutely vital for
all organizations to introduce detectors in their programs and systems which will
alert the user in case any attacks are being made. If the authorities do not focus on
the actual problem that lies within the system, more information breaches will take
place. Attackers these days make use of personal information and end up threatening
their targets for money or some other ransom. Thus, through such researches, the
authorities to take appropriate steps and introduce strong security systems in their
existing or future products.
5 Security Challenges
Flying Ad hoc Networks (FANETs) are highly dynamic, and reliable and offer
multiple services, but with limited access to the network infrastructure. During the
design of FANET architecture, the challenges of security must be considered. Figure 3
shows the type of malicious activity in FANET. Following are some of the challenges:
80
S. Lateef et al.
Fig. 3 Security in FANET
5.1 Dynamic Network Topology
The network topology of FANET is not constant and changes continuously due to
high mobility of aerial vehicles. Thus, it makes the network vulnerable to attacks
and threats [19].
5.2 High Mobility
FANETs have high mobility as aerial vehicles are moving at high speed that may
cause a delay in communication. The aerial vehicles stay connected for short time
and as aerial vehicle move towards its destination it get lost [20].
5.3 Error Tolerance
FANET holds critical information on which actions are performed by aerial vehicles
and infrastructure. Thus, a single mistake in protocols or algorithm can harm the
system badly [21].
Security Threats in Flying Ad Hoc Network (FANET)
81
5.4 Latency Control
Most of the applications of FANET are time sensitive i.e. sending and receiving
messages like collision avoidance, hazard warning and accident warning information
etc. Thus, to achieve real time restraint, cryptographic and other algorithm used in
security must be fast and efficient [17].
5.5 Key Distribution
In FANETs, all algorithms in security are key dependent. Each message is encrypted
or decrypted by same or different key. Thus, key distribution is a main challenge in
designing security protocol [1].
5.6 Data Consistency
In FANET, the information can be altered by any authenticated or unauthenticated
malicious node which can lead to accidents or disturb the network. Hence, mechanism
is needed to avoid data consistency [22].
5.7 Location Awareness
FANETs are completely dependent on GPS or other location-based applications that
can cause nuisance if any error occurs in them [22].
5.8 Need of High Computational Ability
FANET is highly equipped with sensors and computational devices. Real time
computational power helps to obtain current position, speed and direction of vehicle
at any moment of time. The computational ability of these devices is a challenging
issue [22, 23].
82
S. Lateef et al.
5.9 Privacy
In FANET the privacy is a main concern as there is direct communication between
user and aerial vehicle. Thus, it is most important that the information of driver or
location of vehicle do not get disclose [8].
5.10 Routing Protocol
FANET has to provide efficient routing protocol to manage high speed vehicles that
can deliver the messages in given time without any alteration [24, 25].
5.11 Network Scalability
The scale of FANET is increasing rapidly in the world and in return the main problem
that arises with it that there is an absence of global authority existence that governs
the standards of FANET [26].
6 Security Services
Security in FANETs provide the user with responsibility, trust, availability, repudiation, privacy and confidentiality, management of data, access protection, trace-ability
and error detection. The security services of FANET that need more focus and cover
some of the securities are mentioned below [1, 9].
6.1 Availability
The main function of availability is to make sure that the FANET network and other
devices are functional at all times. The information must be available to the authorized
users when it is required, and response time should be fast without delaying it because
it can make the message meaningless if it is received a few minutes late for specific
applications [7].
Security Threats in Flying Ad Hoc Network (FANET)
83
6.2 Confidentiality
This feature maintains the privacy of the user by ensuring that only the supposed
pilots/controllers have data availability and the outside node does not have access to
it until the pilots/controllers fully receives it. The failure of confidentiality is termed
as a breach that, once happened, cannot be remedied [21].
6.3 Data Integrity
This feature defines the integrity of messages by allowing no sources to alter their
content whilst the communication is occurring. It ensures that the message received
at the destination is precisely the same as the sent from the source [19].
6.4 Authentication
Authentication is a crucial step in ensuring the legitimacy of aerial vehicles. It establishes trust among the aerial vehicles. This feature provides verification of aerial
vehicles and ground systems and validation of integrity of information exchange.
It also ensures that all aerial vehicles are right aerial vehicles to communicate in
network [27].
6.5 Non-repudiation
This feature does not allow the receiver and the sender to back-off from transmitting
data in case of any dispute. Only specific authorities with complete authorization are
allowed to identify an aerial vehicle [11, 28].
7 Types of Attackers
In ad hoc environment, different types of attacks are possible especially in air vehicular domain. The impact of these attacks over the system primarily depends over the
intensions of the attackers behind it. Attackers are classified into different categories
according to their intensions to attack as shown in Table 2.
84
S. Lateef et al.
Table 2 Classification of
attackers
Membership
Intentions
Activity
Scope
Internal
Active
Rational
Local
External
Passive
Malicious
Extended
7.1 Basis of Membership
Any authorized or unauthorized node can perform malicious activity in the network.
There are two types of attackers on this basis [29].
7.1.1
Internal Attacker
An internal attack is performed by internal compromised aerial vehicles. It is mostly
the authenticated user of the network [30].
7.1.2
External Attacker
External attacker is the one which has limited ability to attack. It is not the
authenticated user of the network [7].
7.2 Basis of Intention
Any attack that has some objective behind that attack. Following type of attackers
are identified on this basis [31]:
7.2.1
Malicious Attacker
Malicious attacker wants personal gains.
7.2.2
Rational Attacker
Rational attacker has predictable benefits.
Security Threats in Flying Ad Hoc Network (FANET)
85
7.3 Basis of Activity
An attacker is active and makes frequent changes to network or not, the attackers are
classified as [26]:
7.3.1
Active Attacker
Active attackers generate packets or signals. Attacks made by active attackers are
more effective than of passive attackers.
7.3.2
Passive Attacker
Passive attackers only sense the network and do not alter the network.
7.4 Basis of Scope
An attacker can be classified on the basis of their scope [26]:
7.4.1
Local Attacker
Local attackers only have little scope of the aerial vehicle.
7.4.2
Extended Attacker
An extended attacker has control of several entities.
Further, attacks can be classified into five categories [7, 31, 32]:
1.
2.
3.
4.
5.
Monitoring attacks are attacks involving tracking activities.
Social attacks are attacks where unmoral messages are sent in order to badly
affect other pilots/controllers’ emotions.
Timing attacks are attacks where some time slot is added in the original message.
Application attacks are attacks targeting applications providing added service
in FANETs.
Network attacks are attacks affecting the whole networks.
86
S. Lateef et al.
8 Security Threats
Security threats are categorized into different types that are vulnerable to FANET.
The given Fig. 4 shows the classification of threats based on actions:
8.1 Attack on Availability
The lack in availability of FANET service reduces its efficiency. Following are the
attacks and threats on the availability:
8.1.1
DDoS Attack
DDoS attack have serious impact in any network. The attacker accesses to the critical
information of the user by jamming the system of the network in the aerial vehicle.
It can be implemented by inside or outside of FANET [9, 33].
Fig. 4 Classification of threats based on actions
Security Threats in Flying Ad Hoc Network (FANET)
8.1.2
87
Jamming Attack
Attacker use a signal with equal frequency to disrupt the communication channel
using an overpowered signal in equivalent frequency range. It does not follow any
safety alert becoming the most dangerous attack [34].
8.1.3
Malware Attack
To disturb the normal functioning of FANETs, a malware is installed in ground
station, inter-plane and intra-plane. It can cause malfunctioning to other components
[35].
8.1.4
Broadcast Tampering Attack
In this type of attack, attacker inject erroneous messages which leads to the disturbance in network. It results in concealing the correct safety messages to authorized
aerial vehicles [7].
8.1.5
Black Hole Attack
This type of attack is done by a FANET registered user. The malicious node declines
the packets receiving from the network. This disrupts the routing table and prevents
the important message to the recipients due to malicious node, which pretends to
contribute in non-practical event [11].
8.1.6
Grey Hole Attack
A malicious vehicle node drops the packet from some specific node in network and
forwards all the packets to its destination. It is the variant of black hole attack [5].
8.1.7
Greedy Behaviour Attack
In this attack, a malicious aerial vehicle misuses the MAC (Message Authentication
Code) protocol to increase the bandwidth which cost to other users. This results in
collision and traffic overload which produce delay in legitimate services of registered
user [36].
88
8.1.8
S. Lateef et al.
Spamming Attack
The attacker injects large amount of spam messages in the network, in result collision
occurs due to utilization of bandwidth [10].
8.2 Attack on Confidentiality
Confidentiality is one of the most important security requirements in air vehicular
communication. The confidentiality of messages exchanged between aerial vehicles
are vulnerable to following techniques:
8.2.1
Eavesdropping
Eavesdropping is done to attack the confidentiality of the user. As the name suggests,
this kind of attack targets the privacy of the user and aims to get personal information
[19].
8.2.2
Traffic Analysis Attack
The attacker listens to the message transmission and then analyze its frequency and
duration to extract useful and confidential information between aerial vehicles [5].
8.2.3
Man in the Middle Attack
Attacker aims to distribute and share malicious messages between the aerial vehicle,
results in polluting the network by false information. it usually occurs in the middle
of I2I communication [37].
8.2.4
Social Attack
The attacker sends out messages to divert the attention of user. This type of messages
includes advertisements or some immoral messages to get some reaction from the
pilots/controllers. Thus, it affects the flying experience and performance of aerial
vehicle in FANET [22].
Security Threats in Flying Ad Hoc Network (FANET)
89
8.3 Attack on Data Integrity
The data integrity ensures that exchanges data cannot get altered during transmission.
Following are some of the attacks on integrity of data:
8.3.1
Masquerade Attack
The attacker produces false messages which looks like it comes from the authentic
user within the network. The attacker uses false IDs or stolen password to act as
another aerial vehicle [38, 39].
8.3.2
Replay Attack
This is the kind of attack that restricts the user from recognizing hit-and-run incidents.
Here, the attackers benefit by rewinding the previous data transmission [11, 33, 40].
8.3.3
Message Tempering
In this attack, attacker tends to modify, alter, delete or destroying the existing data.
It usually during U2U and U2I communication [8, 41].
8.3.4
Illusion Attack
In this attack. the aerial vehicles receive data from malicious sensors and antennas
that are placed intentionally in the network. It generates traffic warning messages by
using existing flying conditions hence creating an illusion [1, 29, 33].
8.4 Attack on Authentication
Following attacks compromises the authentication/identification of vehicles which
is the basic need of communication:
8.4.1
Timing Attack
FANET aims to deliver the right security message to the user at the right time. In this
security attack, the attacker fits many slots into the message which as a result lead to
receiving messages at the wrong location [9, 17].
90
8.4.2
S. Lateef et al.
Fake Information
The attacker integrates unscripted data into the network. A node interprets the fabricated data and cause mishaps. The information sends by nodes including certificates,
warnings, security messages and identities are not true [4, 18].
8.4.3
Sybil Attack
Attacker creates an illusion of traffics congestion. The intruder maliciously claims
or steals the identities and use those identities to disrupt the functionality of FANET
[5].
8.4.4
GPS Spoofing
In FANET, the position and location of aerial vehicle holds importance which
should be accurate and authentic. Attacker exploits the GPS device to provide false
information to the user [37].
8.4.5
Tunneling
In this attack, attacker creates a path between two distant nodes giving them the idea
that they are neighbors using an extra communication channel called tunnel. It is
similar to wormhole attack [41].
8.4.6
Key/Certificate Replication
The attacker uses the duplicate keys and certificate which are used as a proof
identification which creates uncertainty and make the traffic situation worse [42].
8.4.7
Node Impersonation Attack
This type of attack occurs when the attacker guesses the right ID of registered user.
This attack not only fool the authorized users but also make the innocent user whose
IDs are stolen to be removed from the network [24].
Security Threats in Flying Ad Hoc Network (FANET)
8.4.8
91
Masquerade Attack
The attacker produces false messages which looks like it comes from the authentic
user within the network. The attacker uses false IDs or stolen password to act as
another aerial vehicle [33].
8.4.9
Message Tempering
In this attack, attacker tends to modify, alter, delete or destroying the existing data.
It usually during U2U and U2I communication [29].
8.5 Attack on Non-repudiation
The main threat in repudiation is denial or attempt to denial by a node involved in
communication. IN this attack two nodes have common IDs making them indistinguishable and hence can be repudiated. This feature does not allow the receiver and
the sender to back-off from transmitting data in case of any dispute.
8.5.1
Loss of Event Trace-Ability
In this attack, attacker denies to engage in transmission of sending and receiving
messages in case of any dispute [2].
The given Table 3 shows the FANET security threat classification based on their
object of action.
9 Solution for Security Threats
Where many articles’ studies present the threats and attacks on FANETs, [22] some
other articles also highlight the existing solutions to those attacks. Some of them are
as follows:
9.1 SEAD
Secure and Efficient Ad hoc Distance. SEAD makes use of the hash function
is a single direction to maintain authenticity. The new routes are updated using
destination-sequence number and the previous route data is discarded.
92
Table 3 Classification of
attacks
S. Lateef et al.
Attack
Attack type
Property violation
DDoS
Malicious, active,
insider, network
Availability
Masquerade
Insider, active
Authentication,
data integrity
Tunnelling
Outsider, malicious, Authentication
monitoring
Fake information
Insider
Authentication
Black hole
Outsider, passive
Availability
Social
Insider
Confidentiality
Malware
Insider, malicious
Availability
Man in the middle
Insider, malicious
Confidentiality,
privacy, data
integrity
Illusion
Insider, outsider
Authentication,
data integrity
Sybil
Insider, network
Authentication,
privacy
GPS spoofing
Outsider
Authentication
Eavesdropping
Insider
Data integrity,
confidentiality
Timing attack
Insider, malicious
Data integrity,
authentication
9.2 Ariadne
Ariadne runs over on-demand routing protocol DSR. The symmetric cryptographic
operations mostly make use of this protocol. This system is introduced in the security
system of TESLA.
9.3 RobSAD
Robust method for Sybil Attack Detection, this operation believes that two different
pilots/controllers cannot have identical motion pattern while flying different aerial
vehicles. So, any suspicious activity is identified if two or more nodes have identical
motion trajectories.
Security Threats in Flying Ad Hoc Network (FANET)
93
9.4 ARAN
Authenticated Routing for Ad hoc Network (ARAN), is a protocol in which third
party is present that provide signed certificate to nodes. Each node coming into the
network need to send request certificate to that third party. Public key of third party
is known to all authorized nodes. Asymmetric cryptographic technique is used for
authenticated secure route discovery and timestamps are used for freshness of route.
ARAN works on five steps:
1.
2.
3.
4.
5.
Certification
Authenticated Route Discovery
Authenticated Route Setup
Route Maintenance
Key Revocation.
Route authentication process is done at each step, through addition of sign and
certificate of each intermediate node, thus impersonation problems are solved by
using this protocol.
9.5 SAODV
All routing messages are digitally signed to ensure authenticity and for protection
hop count hash functions are used. In this approach intermediate node cannot send
route reply even if the fresh route is known to them. Through Double Signature this
problem can be solved but it increases the complexity of the system.
9.6 A-SAODV
This protocol is an extension to SAODV that has an experimental feature of adaptive
reply decision. Each intermediate node can decide whether to send reply to source
node or not, depending on the queue length and threshold conditions.
9.7 One Time Cookie
For session management, cookies are assigned per session. But to prevent the system
from session hijacking and theft of SID, this protocol gives the concept of OTC
(one time cookie). OTC generate token for each request and these tokens are tied
to request using HMAC (Hash based Message Authentication Code) to prevent the
re-use of the token.
94
S. Lateef et al.
Table 4 Summary of solutions
Solution
Technique
Attack
SEAD
One way hash function
DoS, routing attack, nodes
impersonation
ARIADNE
Symmetric cryptography, MAC
DoS, replay, routing attack
ECDSA
Elliptical curve parameter, digital
signature
Fake information, node
impersonation
RobSAD
Motion pattern analysis
Sybil attack
ARAN
Cryptographic technique
Replay, impersonation,
eavesdropping
SAODV
Digital signature, hash function
Routing, impersonation, fake
information
A-SAODV
Digital signature, hash function
Routing attack, fake information,
impersonation
One time cookie
Random cookie generation
Session hijacking
The given Table 4 shows the summary of solutions and Fig. 4 shows the
classification of threats based on action.
10 Conclusion
FANET is a rising technology that provides the user with undoubted convenience.
However, users want safety along with convenience which can only be introduced
into the system by implementing secure and safe FANET applications. FANET aims
to ensure safety to the human beings. It does so by providing services of comfort to
the users as well as by broadcasting safe communication among vehicles however,
the messages are broadcasted in an environment that has open access. This makes the
system highly vulnerable to malicious attacks. The security of FANET has become
highly susceptible to malicious attackers who aim to change the contents of different
applications in order to deceive the users and sometimes to gain open access to their
personal information. Nowadays, FANETs are gaining popularity due to the variety
of ubiquitous services they provide to the public. It is of paramount importance to
ensure FANETs security as their future use should not compromise the user safety. We
believe that the researches on FANET must be focused on the security issues faced by
the system, this would require a critical analysis of the internal system of FANET. In
addition to these, the researches must focus on enhancing the authentication schemes
in order to provide a safe communication in FANET [19].
Security Threats in Flying Ad Hoc Network (FANET)
95
References
1. H. Hasrouny, A.E. Samhat, C. Bassil, A. Laouiti, VANet security challenges and solutions: a
survey. Veh. Commun. 7, 7–20 (2017)
2. K. Singh, S. Sharma, Advanced security attacks on vehicular ad hoc network (VANET). Int. J.
Innov. Technol. Exploring Eng. 9(2), 3057–3064
3. R. Kumar, M. Shanmugam, A detailed case study on VANET security requirements, attacks
and challenges. Adv. Model. Anal. B 62(2–4), 48–52 (2019)
4. H. Shafiq, R.A. Rehman, B.S. Kim, Services and security threats in SDN based VANETs: a
survey. Wireless Commun. Mobile Comput. 1–14 (2018)
5. R. Abassi, VANET security and forensics: challenges and opportunities. Wiley Interdisc. Rev.
Forensic Sci. 1, e1324 (2019). http://doi.org/10.1002/wfs2.1324
6. G. Samara, Y. Al-Raba’nah, Security issues in vehicular ad hoc networks (VANET): a survey
(2017)
7. M. Maad Hamdi, L. Audah, S. Abduljabbar Rashid, A. Hamid Mohammed, S. Alani, A. Shamil
Mustafa, A review of applications, characteristics and challenges in vehicular ad hoc networks
(VANETs), in 2020 International Congress on Human-Computer Interaction, Optimization
and Robotic Applications (HORA) (2020), pp. 1–7. http://doi.org/10.1109/HORA49412.2020.
9152928
8. A. Kumar, M. Bansal, A review on VANET security attacks and their countermeasure, in 2017
4th International Conference on Signal Processing, Computing and Control (ISPCC) (2017),
pp. 580–585. http://doi.org/10.1109/ISPCC.2017.8269745
9. M.M. Hamdi, L. Audah, S.A. Rashid, A review of applications, characteristics and challenges
in vehicular ad hoc networks. Carleton University (2020)
10. B. Sharma, M.S.P. Sharma, R.S. Tomar, A survey: issues and challenges of vehicular ad hoc
networks (VANETs), in Proceedings of International Conference on Sustainable Computing in
Science, Technology and Management (SUSCOM), Amity University, Rajasthan, Jaipur, India,
February 26–28, 2019. Available at SSRN: https://ssrn.com/abstract=3363555. http://doi.org/
10.2139/ssrn.3363555
11. C.M.A. Rasheed, S. Gilani, S. Ajmal, A. Qayyum, Vehicular ad hoc network (VANET): a
survey, challenges, and applications (2017)
12. A. Luckshetty, S. Dontal, S. Tangade, S.S. Manvi, A survey: comparative study of applications,
attacks, security and privacy in VANETs, in 2016 International Conference on Communication
and Signal Processing (ICCSP) (2016), pp. 1594–1598. http://doi.org/10.1109/ICCSP.2016.
7754429
13. I. Bekmezci, O.K. Sahingoz, Ş Temel, Flying ad hoc networks (FANETs): a survey. Ad Hoc
Netw. 11(3), 1254–1270 (2013)
14. I. Sumra, P. Sellappan, A. Abdullah, A. Ali, Security issues and challenges in Manet-VanetFanet: a survey. EAI Endorsed Trans. Energy Web 5(17) (2018)
15. A. Chriki, H. Touati, H. Snoussi, F. Kamoun, FANET: communication, mobility models and
security issues. Comput. Netw. 163, 106877 (2019)
16. A. Mukherjee, V. Keshary, K. Pandya, N. Dey, S.C. Satapathy, Flying ad hoc networks: a
comprehensive survey. Inf. Decis. Sci. 569–580 (2018)
17. A.K. Goyal, G. Agarwal, A.K., Tripathi, Network architectures, challenges, security attacks,
research domains and research methodologies in VANET: a survey. Comput. Netw. Inf. Secur.
(19), 37–44 (2019)
18. S. Sharma, A. Kaul, VANETs cloud: architecture, applications, challenges, and issues. Arch.
Comput. Methods Eng. 1–22 (2020)
19. M.S. Sheikh, J. Liang, A comprehensive survey on VANET security services in traffic
management system. Wirel. Commun. Mobile Comput. 1–23 (2019)
20. E. Ajulo, R. Akinyede, O. Adewale, Security threats and privacy issues in vehicular ad hoc
network (VANET): survey and perspective. J. Inf. 4(1), 1–9 (2018)
96
S. Lateef et al.
21. P.R. Krishnan, A.R. Kumar, Security and privacy in VANET: concepts, solutions and challenges,
in 2020 International Conference on Inventive Computation Technologies (ICICT) (2020),
pp. 789–794. http://doi.org/10.1109/ICICT48043.2020.9112535
22. R. Mishra, A. Singh, R. Kumar, VANET security: issues, challenges and solutions, in
International Conference on Electrical, Electronics, and Optimization Techniques (2016),
pp. 1050–1055
23. N. Chaubey, Security analysis of vehicular ad hoc networks (VANETs): a comprehensive study
(2016)
24. S. Sharma, A. Koul, VANETs cloud: architecture, applications, challenges, and issues. Arch.
Comput. Meth. Eng. (2020). https://doi.org/10.1007/s11831-020-09447-9
25. İ Bekmezci, E. Şentürk, T. Türker, Security issues in flying ad hoc networks (FANETS). J.
Aeronaut. Space Technol. 9(2), 13–21 (2016)
26. F. Qu, Z. Wu, F. Wang, W. Cho, A security and privacy review of VANETs. IEEE Trans. Intell.
Transp. Syst. 16(6), 2985–2996 (2015). https://doi.org/10.1109/TITS.2015.2439292
27. A.S. Al Hasan, M.S. Hossain, M. Atiquzzaman, Security threats in vehicular ad hoc networks,
in 2016 International Conference on Advances in Computing, Communications and Informatics
(ICACCI), Jaipur (2016), pp. 404–411
28. A. Quyoom, A. Mir, D. Sarwar, Security attacks and challenges of VANETs: a literature survey.
J. Multimedia Inf. Syst. 7, 45–54 (2020)
29. M.A. Elsadig, Y.A. Fadlalla, VANETs security issues and challenges: a survey. Indian J. Sci.
Technol. 09(28), 1–8 (2016)
30. Z. Lu, G. Qu, Z. Liu, A survey on recent advances in vehicular network security, trust, and
privacy. IEEE Trans. Intell. Transp. Syst. 20(2), 760–776 (2019). https://doi.org/10.1109/TITS.
2018.2818888
31. P. Agarwal, N. Bhardwaj, A review on trust model in vehicular ad hoc network. Int. J. Grid
Distrib. Comput. 9, 325–334 (2016). http://doi.org/10.14257/ijgdc.2016.9.4.29
32. W. Ahmed, M. Elhadef, Securing intelligent vehicular ad hoc networks: a survey, in Advances
in Computer Science and Ubiquitous Computing (Springer, Singapore, 2017), pp. 6–14
33. A. Kumar, M. Bansal, A review on VANET security attacks and their countermeasure, in
International Conference on Signal Processing, Computing and Control (2017), pp. 580–585
34. A. Kchaou, R. Abassi, S. Guemara, Toward a distributed trust management scheme for vanet,
in Proceedings of the 13th International Conference on Availability, Reliability and Security
(2018), pp. 1–6
35. M.N. Javed, H. Shafiq, K.A. Alam, A. Jamil, M.U. Sattar, VANET’s security concerns and
solutions: a systematic literature review, in Proceedings of the 3rd International Conference
on Future Networks and Distributed Systems (2019), pp. 1–12
36. M. Abu Talib, S. Abbas, Q. Nasir, M.F. Mowakeh, Systematic literature review on internetof-vehicles communication security. Int. J. Distrib. Sens. Netw. 14(12), 1550147718815054
(2018)
37. Z. Afzal, M. Kumar, Security of vehicular ad hoc networks (VANET): a survey. J. Phys. Conf.
Ser. 1427, 012015 (2020). http://doi.org/10.1088/1742-6596/1427/1/012015
38. S. Sharma, B. Kaushik, A survey on internet of vehicles: applications, security issues &
solutions. Veh. Commun. 20, 100182 (2019)
39. Z. Lu, Q. Wang, G. Qu, H. Zhang, Z. Liu, A blockchain-based privacy-preserving authentication
scheme for vanets. IEEE Trans. Very Large Scale Integr. (VLSI) Syst. 27(12), 2792–2801 (2019)
40. R. Abassi, VANET security and forensics: challenges and opportunities. Forensic Sci. 1–13
(2018)
41. R. Mishra, A. Singh, R. Kumar, VANET security: issues, challenges and solutions, in 2016
International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT)
(2016), pp. 1050–1055. http://doi.org/10.1109/ICEEOT.2016.7754846
42. M. Saini, H. Singh, VANET its characteristics attacks and routing techniques: a survey. Int. J.
Sci. Res. 5(5), 1595–1599 (2016)
Secure Communication Routing
in FANETs: A Survey
Shaheen Ahmad and Muhammad Abul Hassan
Abstract Flying ad hoc networks are quite flexible due to the dynamic nature. UAVs
freely move in three-dimensional environment. Wireless networks like 5G enhance
the capabilities of UAV-networks. However, integration of cellular-web of networks
with UAVs has extensive applications. Physical structure of FANETs consists of
aerial vehicles, base station and satellite which collaborate and coordinate with
each system. Due to routing UAV nodes exchange information which cause energy
consumption. This paper presents a comprehensive survey to cover different areas
in flying ad hoc networks which include routing protocols, wireless technologies
and mobility patterns. Apart from that MANET, VANET and FANETs are discussed
properly with respect to different metrics. In addition, IoFN is formulated which
is combination of IoT FANETs to improve communication and reduce end-to-end
delay. Also, limitation regarding the mentioned areas in FANETs are incorporated.
Keywords Flying ad hoc networks · Secure routing protocols · Mobility models ·
Unmanned aerial vehicles · Mobile ad hoc network · Vehicular ad hoc network
1 Introduction
The internet of flying networks is a hybrid of two popular technologies: the internet
of things and flying ad hoc networks. Mostly on ground, IoT sensor devices provide
information to aerial vehicles. Flying-IoT forms a network in the sky that collects
data from ground nodes and updates the base station. Unmanned aerial vehicles
(UAVs) are considered as the most promising technology with many civil and military applications [1, 2]. Most countries rely on agriculture, where smart farming
is one of the best approaches. To check the water level, IoT-based sensor nodes
are embedded in trees and plants. Aerial vehicles are being employed to transport
medical instruments and medicines to remote locations. UAVs are incredibly helpful
in surveillance and monitoring, which may be utilized to ensure humanity’s safety.
S. Ahmad (B) · M. A. Hassan
Department of Computing and Technology, Abasyn University, Peshawar, Khyber Pakhtunkhwa
25000, Pakistan
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_6
97
98
S. Ahmad and M. A. Hassan
UAV enabled networks plays an important role in tracking and monitoring players
specially in games. Wireless body area sensor networks in combination with UAVs
collect information and send directly to land station. Aerial vehicles are having
direct role in vaccine supply, public temperature measurement, mask monitoring
and first aid supply. Therefore, protocols can be used for optimal communication
within network [3]. Routing protocols are used to provide effective communication
between aerial vehicles. However, since this field of study is relatively new, MANET
protocols were first employed, which posed a significant difficulty. Researchers and
scientists began working on novel routing protocols for flying ad hoc networks as a
result of these findings. Although energy is seen as a key issue, E-AntHocNet was
designed with the goal of saving energy and extending battery life in aerial vehicles. This protocol basic working is based on ant colony optimization, where a new
parameter is implemented named energy stabilizing threshold [4]. Cognitive dynamic
source routing strategy is proposed which select optimal route selection from source
to target. Broadcasting and data transmission are most well-known feature in DSR
routing. Channel route failure, mobility and end to end delay is having direct effect on
IoT based networks. On ground IoT and software define networks are directly interconnected with FANETs. Therefore, cognitive IoT concept is introduced to explore
big data-based routing techniques to select possible route selection in network [5].
Bluetooth technology is being utilized for the first time in MANETs and VANETs.
As WLAN technology advanced with the passage of time Zigbee is launched. In
the main topological structure of flying ad hoc networks consist of base station,
aerial vehicles and satellite. Therefore, 3-D centroid algorithm was used to improve
signal power in wireless sensor networks [6]. However, finding actual, location of
aerial vehicles decision tree classifier improves received signal strength [7]. Wireless
technology is considered the backbone approach in flying ad hoc networks. A brief
survey is conducted from 1 to 6G. Also, protocols like AntHocNet, ZRP, M-DART,
AOMDV, DSR, DSDV are simulated using network simulator-2 [8].
Communication networks are quite vulnerable to security attacks include
DoS/DDoS and ping of death. The intruder tries to send multiple data packets using
flooding technique to spoof the nodes. Delay is caused due to attacks where datagram transport layer security is used to secure two nodes. Therefore, intrusion detection system is application that directly detects different attacks. IDS is categorized
into anomaly, signature and hybrid. The criteria of false alarm are detected easily
by checking queue lengths in networks [9]. Due to the mentioned reason securitybased routing technique should be designed in flying ad hoc networks. The existing
routing protocols of mobile ad hoc networks should be redesign specially for internet
of drones. Flying-IoT based networks having dynamic nature due to which implementation of routing protocols are very tough. Multi-point relay is the technique
use to overcome on congestion. Random walk mobility model is used to check the
performance of FANETs routing protocols [10]. Figure 1, discuss different real-time
applications for UAVs like public health, examination, electric city, goods supply and
lab test monitoring. However, Table 1 represents MANETs, VANETs and FANETs
comparison with different parameters.
Secure Communication Routing in FANETs: A Survey
99
Fig. 1 Unmanned aerial vehicle applications
Table 1 Comparisons of MANET, VANET and FANET
MANET
VANET
FANET
Mobility of nodes
Slow
Fast
Very fast
Mobility models of
nodes
Based on random
patterns
Based on regular
patterns
Regular for static and
autonomous for
multiple unmanned
aerial networks system
Network density
(nodes)
Low
High
Based on requirement
Topology reshaping
Slow due less speed
High due high speed Very high due to high
speed
Radio communication Mobile on ground and
loss of signal is
possible in any
situations
Loss of signal is
possible in all cases
Operate in air, but loss
of signal is still existing
Power and network
surveillance time
Energy efficient
protocols are available
Not required due to
constant supply of
power
Energy efficient
protocols are available
for maximum network
surveillance
Consumption on
computation (power)
Very limited
High
High
Nodes localization
GPS
GPS, AGPS, DGPS
IMU, GPS, AGPS,
DGPS
100
S. Ahmad and M. A. Hassan
2 Literature Review
The literature study explains the drawbacks of different routing protocols in flying
ad hoc networks. The behavior of internet of flying drones uses routing protocols for
efficient communication within networks [11, 12]. Therefore, fisheye state routing
is introduced for the first time in the dynamics of UAV-networks. FSR is based
on proactive approach with constantly updates to compute optimal path. During
experimentation of FSR with other contemporary routing protocols, some limitations
has been observed which include link breaks, temporary loops and re-computation
of routing tables is found which directly affect the performance [13].
The existing routing protocols was earlier used in mobile ad hoc networks then
later on researchers implemented the similar schemes but they faced overhead and
broadcasting issues [17, 19, 20]. DSR, AODV, TORA are demonstrated with optimized link state routing protocol. Where, OLSR has shown better results in comparison with other techniques. Multi-point relay is utilized to overcome on overhead
problems. UAVs are divided in clusters and every cluster head is considered MPR
which only do broadcast data packets to the neighbor nodes [24].
Flying ad hoc networks are having better communication channels by using
6G technology for interconnectivity. Aerial vehicles can be used in many different
fields like recuse operations, forestry, medical equipment’s delivery and many more
[15–18]. Therefore, Antonio Guillen-Perez et al., deployed OLSR, P-OLSR and
BATMAN-ADV to check the network bandwidth and throughput for evaluating
adaptability [25]. Figure 2, shows the idea of MPR approach in flying ad hoc
networks.
Fig. 2 MPR technique in FANETs
Secure Communication Routing in FANETs: A Survey
101
Qianqian Sang et al., conducted a survey on routing protocols highlighting the
key areas like energy efficiency, scalability, survivability and accuracy is discussed
with respect to flying ad hoc networks [22, 23, 26]. A seven years study is examined to evaluate network architecture, design, routing techniques, quality of service
issues and some major issues are well explained. However, with routing protocols
deployment of mobility models are very much important to avoid collision [27]. UAVnetworks are divided into two main categories which consist of single and multiple
UAV structure. Some design aspects of routing must have better delivery, cooperative
routing, path discovery and prediction. Some well-known clustering techniques are
used in flying-IoT networks. However, issues like link disconnection, hybrid metrics,
performance awareness and security are finalized to improve performance [3, 28].
UAV-enabled IoT based networks has transforms the structure of health and sports.
Inam Ullah Khan et al., has deployed Anthocnet, AOMDV, DSDV, DSR, M-DART
and ZRP. The suggested protocols have given solution to loop free environment,
localization and better quality of service in flying networks [29]. In addition, cellular
networks are not available everywhere, therefore mobile nodes are utilized to enable
data packets among vehicular delay tolerant networks. The proposed solution easily
overcome on end-to-end delay and reduced network overhead [30]. Apart from that
flocking based on demand routing protocol is designed for aerial networks. A novel
routing protocol Boids of Reynolds mechanism is used in AODV protocol to maintain
routes. The solution has improved delay, throughput and packet loss [31]. Table 2,
illustrates the past researchers work regarding UAVs.
3 Wireless Communication
Initially first-generation cellular communication was introduced in 1980’s. Therefore, second-generation expanded with small cell coverage with digital technology.
Then later Pico cellular approach came with the understanding of base station. Thirdgeneration given the concept of CDMA technologies in year 2000. Where, 3G was
quite fast in transmission, location services and connecting medical datasets through
wireless means internet technology [32]. Apart from that 4G communication 1 Gbps
per device peak data rate. The latency is quite high about 100 ms [33]. There is no
satellite integration and mobility range are hardly 350 km/h. However, 5G networks
significantly improves the existing systems [34]. For, ultra-long-range communication 6G networks is used which support several new technologies [1, 35]. 6G need to
satisfy physical layer security as well. Authentication use to public key cryptography
to ensure security counter measure [36]. Table 3, shows 3G to 6G comparative analysis. While Fig. 3, elaborates the concept of smart cities. Therefore, aerial vehicles
improve communication channels in smart city.
UAVs classification
Not provided
Not provided
Not provided
Not provided
Not provided
Provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Refs.
[14]
[1]
[2]
[15]
[16, 17]
[18]
[17, 19]
[20]
[21]
[22]
[11, 23]
[12]
Not provided
Not provided
Provided
Not provided
Provided
Not provided
Provided
Provided
Provided
Not provided
Provided
Provided
Communication
architecture of
UAVs
Table 2 Comparison of different mobility models
Not provided
Not provided
Partially
provided
Not provided
Not provided
Not provided
Provided
Provided
Not provided
Provided
Provided
Provided
Applications of
UAVs
Provided
Not provided
Not provided
Partially provided
Not provided
Not provided
Partially provided
Not provided
Not provided
Provided
Not provided
Not provided
Mobility models
Not provided
Provided
Not provided
Not provided
Provided
Provided
Provided
Not provided
Not provided
Not provided
Not provided
Not provided
Routing taxonomy
Provided to only
position based
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Partially provided Not provided
Not provided
Not provided
Partially provided Not provided
Partially provided Provided
Provided to only
position based
Not provided
Not provided
Not provided
Partially provided Not provided
Partially provided Not provided
Routing protocols Comparative
study
102
S. Ahmad and M. A. Hassan
Secure Communication Routing in FANETs: A Survey
103
Table 3 3G to 6G comparative analysis
Technology
Practically
deployed
Bandwidth
(Data)
Multiplexing
Network
Core N/W
Third
Generation
(3G) [32]
2004–2010
2 Mbps
CDMA
Packet
Packet N/W
Fourth
Generation
(4G) [33]
2010
1 Gbps
CDMA
Hybrid
Internet
Fifth
Generation
(5G) [34]
2015
Higher then
1 Gbps
CDMA
Hybrid
Internet
Sixth
Generation
(6G) [35]
2020
70 GB
CDMA
Hybrid
IPV6
satellite-based
connectivity
Fig. 3 Smart city with technological fields
104
S. Ahmad and M. A. Hassan
3.1 Mobility Models
UAV-networks perform all sort of tasks which include monitoring and surveillance.
UAV swarm use to have so many advantages over single UAV. Aerial vehicles need
data transmission among each UAV and base station. While facing communication
problems routing protocols try to improve channels. UAVs move in three-dimensional
environment. Therefore, each node needs to be properly positioned to check FANETs
performance. However, due to the mentioned issues flying ad hoc networks need
mobility models [21, 37]. Some mobility models are discussed as under.
3.1.1
Random Walk
To detect unpredictable movements of nodes random walk is introduced. As, this
mobility model is considered memory less because it doesn’t have the capability to
store past speed, position and direction [14].
3.1.2
Random Way Point
Random way point is so much similar with RW but having some differences. This
mobility model is time dependent, if time will be covered then mobile node chooses
random destination in simulation area [38].
3.1.3
Manhattan Grid
For road topological scenario Manhattan grid mobility model is proposed specially
for urban areas. This pattern uses probabilistic approach for flying nodes movement
in FANETs [39, 40].
3.2 Time-Dependent Mobility Models
There are two more types of mobility patterns which include:
3.2.1
Boundless Simulation Area
BSA mobility pattern use the relationship between past and current mobile node
direction, speed and movements. In this mobility model effective edge is removed
in simulation evaluation [39].
Secure Communication Routing in FANETs: A Survey
3.2.2
105
Gauss-Markov
This mobility model uses to adapt the randomness. GM usually able to reduce on
the spot stops and sharp turns [40]. Table 4 represents overall mobility models’
comparison.
3.3 Routing Protocols in FANETs
Routing protocols played important role in Flying ad hoc Networks.
3.3.1
I-BAT Coop Protocol
The environment of flying ad hoc networks is quite challenging. The dynamic structure and node movement focus on multi-path transmission. Many issues have been
arising in FANETs which can be easily overcome using game theory to clustering
channel estimation. Therefore, biologically inspired technique called BAT algorithm
is improved with using multiple relays. iBAT-Coop protocol is specially designed
for FANETs to reduce packet loss, end-to-end delay and transmission loss can be
minimized about 82%. The proposed solution is compared with BAT-FANET. In
addition, average link duration is improved about 25% in iBAT-Coop [27].
3.3.2
MPEAOLSR Protocol
Improved version of OLSR is take into consideration which is called multidimensional perception and energy awareness-OLSR. However, multi-point relay
is used to improve broadcasting overhead problems. Link layer congestion level and
path selection is optimized well in FANETs. Quality of service parameters like packet
loss, delay and data transmission rate are enhanced in flying networks [41].
3.3.3
Dynamic Dual Reinforcement Learning Routing
Quality of experience-based management for multi-service flying mesh networks are
investigated. Backward/Forward agent approach is utilized to dynamically improve
routing protocol. Therefore, reinforcement learning is combined with many routing
protocols through which UAV-network can directly learn from environment [42].
Time
based
mobility
models
Random
based
mobility
models
Not supported
Not supported
Nodes randomness Not supported
is supported
Smooth turn [40]
Not supported
Gauss-Markov [39]
Nodes can select
random direction
Random walk [39]
Not supported
Not supported
After some period
of time nodes
selects random
direction
Random way point
[38]
Not supported
Nodes collision
avoidance
Boundless simulation Nodes selects
area [40]
direction after
some specific
period of time
Able to select
random direction
in given space
Random selection
Manhattan grid [37]
Mobility model
Table 4 Mobility models comparison
Dimensions
Utilized in real
application
Not supported Three-dimensional
area
Not supported Three-dimensional
area
Not supported Three-dimensional
area
Target finding
missions
Unpredictable
movements mission
Can be deployed in
any situation
Not supported Two dimension area Random movements
space
of fights
Not supported Two dimension area Mostly used in
space
patrolling. e.g.
boarder, agriculture
usage
Not supported Two dimension area Traffic patrolling
space
Nodes
connectivity
Randomly opt
direction with
fixed turn radius
Future movement
is predicted based
on previous ones
Selects boundless
pattern
Speed and
directions are
select on random
bases
Each node has
pause time
waiting
mechanism to
select another
possible random
direction
Track down
vehicle
movements in
rush hours
Working
106
S. Ahmad and M. A. Hassan
Secure Communication Routing in FANETs: A Survey
3.3.4
107
Destination Sequence Distance Vector Routing
DSDV routing protocol is based on bellman Ford algorithm. DSDV is basically proactive which incrementally update data packets. Network topology is use to change
accordingly and reduce overhead problems. Also, network loops are properly avoided
and high bandwidth is used to update the path [43].
3.3.5
Radio-Metric AODV
RM-AODV is proposed using IEEE 802.11s. The protocol is designed to work at layer
2 where MAC protocol is on layer 3. This approach tries to improve and maintain
path discovery process [44].
3.3.6
Temporary Ordered Routing Algorithm
TORA protocol uses link reversal which maintain directed graph from source to
target. TORA use to minimize network load and optimize shortest path [45]. Figure 4,
routing protocols in FANETs which can be either proactive, reactive or hybrid.
Fig. 4 Routing protocols in FANETs
108
S. Ahmad and M. A. Hassan
4 Conclusion
In the area of flying ad hoc networks routing use to improve the main features
and core network operations. During the last few decades researchers developed
many routing protocols in FANETs. Every protocol uses to have various characteristics, limitations and advantages. Comprehensive study regarding FANETs communication is discussed to improve the behavior of UAVs. FANETs originally taken
idea from MANETs where mobility models play an important role in effectiveness of network. This paper, shows UAV study which contains routing protocols,
mobility patterns and wireless technologies. Also, MANET, VANET and FANET
are compared with respective parameters. However, this research study is helpful for
engineers, scientists, researchers and technical practitioners.
References
1. A. Bujari, C.T. Calafate, J.-C. Cano, P. Manzoni, C.E. Palazzi, D. Ronzani, Flying ad hoc
network application scenarios and mobility models. Int. J. Distrib. Sens. Netw. 13(10),
1550147717738192 (2017)
2. J. Li, Y. Zhou, L. Lamont, Communication architectures and protocols for networking
unmanned aerial vehicles, in Proceedings IEEE Globecom Workshops (GC Wkshps), Atlanta,
GA, USA (2013), pp. 1415–1420
3. I.U. Khan, M.A. Hassan, M.D. Alshehri, M.A. Ikram, H.J. Alyamani, R. Alturki, V.T. Hoang,
Monitoring system-based flying IoT in public health and sports using ant-enabled energy-aware
routing. J. Healthc. Eng. 2021 (2021)
4. U. Khan, I.M. Qureshi, M.A. Aziz, T.A. Cheema, S.B.H. Shah, Smart IoT control-based nature
inspired energy efficient routing protocol for flying ad hoc network (FANET). IEEE Access 8,
56371–56378 (2020). https://doi.org/10.1109/ACCESS.2020.2981531
5. S. Begum, Y. Nianmin, S.B.H. Shah, A. Abdollahi, I.U. Khan, L. Nawaf, Source routing for
distributed big data-based cognitive internet of things (CIoT). Wirel. Commun. Mobile Comput.
2021 (2021)
6. S.B. Shah, C. Zhe, F. Yin, I.U. Khan, S. Begum, M. Faheem, F.A. Khan, 3D weighted centroid
algorithm & RSSI ranging model strategy for node localization in WSN based on smart devices.
Sustain. Cities Soc. 39, 298–308 (2018)
7. I.U. Khan, R. Alturki, H.J. Alyamani, M.A. Ikram, M.A. Aziz, V.T. Hoang, T.A. Cheema,
RSSI-controlled long-range communication in secured IoT-enabled unmanned aerial vehicles.
Mob. Inf. Syst. 2021 (2021)
8. I.U. Khan, S.Z.N. Zukhraf, A. Abdollahi, S.A. Imran, I.M. Qureshi, M.A. Aziz, S.B.H. Shah,
Reinforce based optimization in wireless communication technologies and routing techniques
using internet of flying vehicles, in The 4th International Conference on Future Networks and
Distributed Systems (ICFNDS) (2020), pp. 1–6
9. A. Abdollahi, M. Fathi, An intrusion detection system on ping of death attacks in IoT networks.
Wirel. Pers. Commun. 1–14 (2020)
10. I.U. Khan, S.B.H. Shah, L. Wang, M.A. Aziz, T. Stephan, N. Kumar, Routing protocols &
unmanned aerial vehicles autonomous localization in flying networks. Int. J. Commun. Syst.
e4885 (2021)
11. J. Xie, Y. Wan, J.H. Kim, S. Fu, K. Namuduri, A survey and analysis of mobility models for
airborne networks. IEEE Commun. Surveys Tuts. 16(3), 1221–1238 (2014)
Secure Communication Routing in FANETs: A Survey
109
12. S. Hameed, Q.-A. Minhas, S. Ahmed, S. Habib, M.K. Hasan, M. Islam, S. Khan, An improved
iBAT-COOP protocol for cooperative diversity in FANETs. CMC-Computers Mater. Continua
67(2), 2527–2546 (2021)
13. M.A. Hassan, S.I. Ullah, I.U. Khan, S.B.H. Shah, A. Salam, A.W.U. Khan, Unmanned aerial
vehicles routing formation using fisheye state routing for flying ad hoc networks, in The 4th
International Conference on Future Networks and Distributed Systems (ICFNDS) (2020),
pp. 1–7
14. L. Gupta, R. Jain, G. Vaszkun, Survey of important issues in UAV communication networks.
IEEE Commun. Surveys Tuts. 18(2), 1123–1152 (2016)
15. S. Hayat, E. Yanmaz, R. Muzaffar, Survey on unmanned aerial vehicle networks for civil
applications: a communications viewpoint. IEEE Commun. Surveys Tuts. 18(4), 2624–2661
(2016)
16. V. Sharma, R. Kumar, Cooperative frameworks and network models for flying ad hoc networks:
a survey. Concurrency Comput. Pract. Experience 29(4) (2017)
17. O.S. Oubbati, A. Lakas, F. Zhou, M. Gunes, M.B. Yagoubi, A survey on position-based routing
protocols for flying ad hoc networks (FANETs). Veh. Commun. 10, 29–56 (2017)
18. J. Jiang, G. Han, Routing protocols for unmanned aerial vehicles. IEEE Commun. Mag. 56(1),
58–63 (2018)
19. M.H. Tareque, M.S. Hossain, M. Atiquzzaman, On the routing in flying ad hoc networks, in
Proceedings of IEEE Federated Conference on Computer Science and Information Systems
(FedCSIS), Lodz, Poland (2015), pp. 1–9
20. M.A. Khan, A. Safi, I.M. Qureshi, I.U. Khan, Flying ad hoc networks (FANETs): a review
of communication architectures, and routing protocols, in Proceedings of 1st IEEE International Conference on Latest Trends in Electrical Engineering and Computing Technologies
(INTELLECT), Karachi, Pakistan (2017), pp. 1–9
21. K. Kumari, B. Sah, S. Maakar, A survey: different mobility model for FANET. Int. J. Adv. Res.
Comput. Sci. Softw. Eng. 5(6) (2015)
22. K. Namuduri, Y. Wan, M. Gomathisankaran, Mobile ad hoc networks in the sky: state of the art,
opportunities, and challenges, in Proceedings of 2nd ACM MobiHoc Workshop on Airborne
Networks and Communications (2013), pp. 25–28
23. O.K. Sahingoz, Networking models in flying ad hoc networks (FANETs): concepts and
challenges. J. Intell. Rob. Syst. 74(1–2), 513 (2014)
24. M.A. Hassan, S.I. Ullah, A. Salam, A.W. Ullah, M. Imad, F. Ullah, Energy efficient hierarchical
based fish eye state routing protocol for flying Ad hoc networks. Indonesian J. Electr. Eng.
Comput. Sci. 21(1) 465–471 (2021)
25. A. Guillen-Perez, A.-M. Montoya, J.-C. Sanchez-Aarnoutse, M.-D. Cano, A comparative
performance evaluation of routing protocols for flying ad hoc networks in real conditions.
Appl. Sci. 11(10), 4363 (2021). Q. Sang, H. Wu, L. Xing, P. Xie, Review and comparison of
emerging routing protocols in flying ad hoc networks. Symmetry 12(6), 971 (2020)
26. X. Fan, W. Cai, J. Lin, A survey of routing protocols for highly dynamic mobile ad hoc networks,
in Proceedings of 17th IEEE International Conference on Communication Technology (ICCT),
Beijing, China (2017), pp. 1412–1417
27. M.Y. Arafat, S. Moh, Routing protocols for unmanned aerial vehicle networks: a survey. IEEE
Access 7, 99694–99720 (2019)
28. A. Rovira-Sugranes, F. Afghah, J. Chakareski, A. Razi, A review of AI-enabled routing protocols for UAV networks: trends, challenges, and future outlook. arXiv preprint arXiv:2104.
01283 (2021)
29. Z. Du, C. Wu, T. Yoshinaga, X. Chen, X. Wang, K.-L.A. Yau, Y. Ji, A routing protocol for
UAV-assisted vehicular delay tolerant networks. IEEE Open J. Comput. Soc. 2, 85–98 (2021)
30. N. El Houda Bahloul, S. Boudjit, M. Abdennebi, D.E. Boubiche, A flocking-based on demand
routing protocol for unmanned aerial vehicles. J. Comput. Sci. Technol. 33(2), 263–276 (2018)
31. I. Bekmezci, O.K. Sahingoz, Ş. Temel, Flying ad hoc networks (FANETs): a survey. Ad Hoc
Netw. 11(3), 1254–1270 (2013)
110
S. Ahmad and M. A. Hassan
32. Y. Zhao, Standardization of mobile phone positioning for 3G systems. IEEE Commun. Mag.
40, 108–116 (2002)
33. M.A. Spirito, On the accuracy of cellular mobile station location estimation. IEEE Trans. Veh.
Technol. 50, 674–685 (2001)
34. F. Bai, N. Sadagopan, A. Helmy, The important framework for analyzing the impact of mobility
on performance of routing protocols for ad hoc networks. Ad Hoc Netw. 1(4), 383–403 (2003)
35. J. Yoon, M. Liu, B. Noble, Random waypoint considered harmful, in Proceedings of the TwentySecond Annual Joint Conference of the IEEE Computer and Communications (INFOCOM’03),
vol. 2 (IEEE, 2003), pp. 1312–1321
36. J.A. del Peral-Rosado et al., Survey of cellular mobile radio localization methods: from 1G to
5G. IEEE Commun. Surv. Tutorials 20(2), 1124–1148 (2017)
37. K.-H. Chiang, N. Shenoy, A 2-D random-walk mobility model for location-management studies
in wireless networks. IEEE Trans. Veh. Technol. 53(2), 413–424 (2004)
38. T. Camp, J. Boleng, V. Davies, A survey of mobility models for ad hoc network research. Wirel.
Commun. Mob. Comput. 2(5), 483–502 (2002)
39. V. Tolety, Load reduction in ad hoc networks using mobile servers (1999)
40. Y. Wan, K. Namuduri, Y. Zhou, S. Fu, A smooth-turn mobility model for airborne networks.
IEEE Trans. Veh. Technol. 62(7), 3359–3370 (2013)
41. C.J. Katila, A. Di Gianni, C. Buratti, R. Verdone, Routing protocols for video surveillance
drones in IEEE 802.11s wireless mesh networks, in 2017 European Conference on Networks
and Communications (EuCNC) (IEEE, 2017), pp. 1–5
42. E. Felemban, Evaluation of routing protocols and mobility in flying ad hoc network. (IJACSA)
Int. J. Adv. Comput. Sci. Appl. 12(7) (2021)
43. A.C. Chen, Advances in wireless communications technologies and their potential biomedical
applications, in Proceedings 1998 IEEE International Conference on Information Technology
Applications in Biomedicine, ITAB’98 (Cat. No. 98EX188) (IEEE, 1998), pp. 82–84
44. M.Z. Chowdhury, M. Shahjalal, S. Ahmed, Y.M. Jang, 6G wireless communication systems:
applications, requirements, technologies, challenges, and research directions. IEEE Open J.
Commun. Soc. 1, 957–975 (2020)
45. Y. Zhao, W. Zhai, J. Zhao, T. Zhang, S. Sun, D. Niyato, K.-Y. Lam, A comprehensive survey
of 6g wireless communications. arXiv preprint arXiv:2101.03889 (2020)
Impact of Routing Techniques
and Mobility Models on Flying Ad Hoc
Networks
Muhammad Abul Hassan, Muhammad Imad, Tayyabah Hassan,
Farhat Ullah, and Shaheen Ahmad
Abstract In wireless sensor networks, nodes can be mobile or static depending
upon application. Unmanned Aerial Vehicle (UAV) is composed of wireless sensors.
UAVs are connected on temporary bases creating Flying Ad hoc Network (FANET).
It is considered as sub-class of Mobile Ad hoc Network (MANET) and Vehicular
Ad hoc Network (VANET). Ease of deployment and hardware cost makes FANET
best choice over other Ad hoc Networks. Mobility Models and Routing Techniques
are two main components in deployment of FANET for maximum output. In this
research study we have studied all Routing Techniques and Mobility Models which
are utilized in various applications of FANET. Finally, we suggest best suitable
mobility models and routing techniques for each FANET application.
Keywords Flying ad hoc networks · Routing techniques · Mobility models ·
Unmanned aerial vehicles · Mobile ad hoc network · Vehicular ad hoc network
1 Introduction
Networks which enabled to share resources to its connected devices e.g. printers,
scanners and even virtual space. Networking appeared in mid-1960 developed by
Advanced Research Agency (ARPANET) for the purpose of to share secret military information among connected devices. Later on, another networking mechanism is developed for public usage called National Science Federation Network
(NSFNET) in mid-1980 as shown in Fig. 1 [1]. So, in Internet now days we are using
M. A. Hassan (B) · M. Imad · S. Ahmad
Department of Computing and Technology, Abasyn University, Peshawar 25000, Khyber Pakhtun
Khwa, Pakistan
T. Hassan
Department of Computer Sciences, Kinnaird College for Women Lahore, Punjab 54000, Pakistan
F. Ullah
School of Automation, Control Sciences and Engineering, China University of Geosciences,
Wuhan 430074, China
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_7
111
112
M. A. Hassan et al.
Fig. 1 Evaluation of internet
is evolved from ARPANET and NSFNET which allow different type of devices to
share information with each other.
1.1 Types of Networks
There are nine types of network we are using now a days are listed below.
1.1.1
Personal Area Network (PAN)
Personal Area Network (PAN) is an innovation that could empower wearable PC
gadgets to communicate with other. For instance, two individuals each wearing business card-size transmitters and recipients possibly could trade data by shaking hands.
The transference of information through intra-body contact by handshakes, is known
as linkup. The human body’s characteristic saltiness makes it a decent conduit of
power. An electric field passes little flows, known as Pico amps, through the body
when the two individuals shake hands. The handshake finishes an electric circuit and
every individual’s information, for example, email locations and telephone numbers,
are moved to the next individual’s PC a comparable gadget. An individual’s garments
likewise could go about as a system for moving this information [2].
1.1.2
Local Area Network (LAN)
Local Area Network (LAN) is an assortment of connected devices in one physical
area, for example, a building, office, or home. Local Area Network can be small or
Impact of Routing Techniques and Mobility …
113
large based on topology of network, running from a home system with one client to
an undertaking system with multiple number of clients and devices working area [3].
1.1.3
Wireless Area Network (WLAN)
Working like a LAN, WLANs utilize remote system innovation, for example, Wi-Fi.
Ordinarily found in indistinguishable sorts of utilizations from LANs, these kinds
of systems don’t necessitate that devices depends on physical links to communicate
with the system [4].
1.1.4
Campus Area Network (CAN)
This type of network is larger than LAN and smaller then Metropolitan Area Network
(MAN). It connects two or three building or K-12 districts and can be installed which
is relatively close to each other [5].
1.1.5
Metropolitan Area Network (MAN)
This type of network is much larger than LAN but smaller than Wide Area Network.
MAN is installed in one close geographic area building like connecting one city
building with each other [6].
1.1.6
Wide Area Network (WAN)
Lager then LAN and MAN and connects devices which is physically far away from
each other basic example is internet which connects two distant devices [7].
1.1.7
Storage Area Network (SAN)
As a committed fast speed that associates shared pools of capacity storage to a few
servers, these kinds of systems don’t depend on a LAN or WAN. Rather, they move
stockpiling assets from the system and spot them into their own high-performance
systems. SANs can be gotten to in a similar manner as a drive connected to a server.
Sorts of capacity region systems incorporate joined, virtual and brought together
SANs [8].
114
1.1.8
M. A. Hassan et al.
System Area Network (Known as SAN)
This term is genuinely new inside the previous two decades. It is utilized to clarify a
generally nearby system that is intended to give rapid association in server-to-server
applications (bunch situations), stockpiling territory systems (called “SANs” too)
and processor-to-processor applications. The PCs associated on a SAN work as a
single system at high speeds [9].
1.1.9
Passive Optimal Area Network (POLAN)
As an option in contrast to customary switch-based Ethernet LANs, POLAN innovation can be coordinated into organized cabling to conquer concerns about supporting
conventional Ethernet conventions and system applications, for example, Power over
Ethernet (PoE). A point-to-multipoint LAN design, POLAN utilizes optical splitters
to part an optical sign from one strand of single mode optical fiber into various signs
to serve clients and devices [10]. Different types of networks are presented in Fig. 2.
1.2 Traditional Network
Data center networks framework for enormous organizations and large compute
farms was assembled dependent on a three-layer various leveled (hierarchical)
model. This conventional system engineering is otherwise called a three-level design,
Fig. 2 Different types of
networks
Impact of Routing Techniques and Mobility …
115
Fig. 3 Switching architecture of network
which Cisco, calls the “various leveled between systems administration model. “It
comprises of center layer switches which associate with circulation layer switches
(once in a while called total switches), which thusly interface with get to layer
switches ($). Access layer switches are often situated at the highest point of a rack,
so they are otherwise called top of rack (ToR) switches as shown in Fig. 3 [11].
Problem associated with this network is its expensive hardware and it is not deterministic due to these constrains for that reason networks are shifted from traditional
network to Ad hoc networks.
2 Background Study
Ad hoc networks are associated with temporary bases or which provide services for
specified situations. Ad hoc networks are sub-divided in to categories based on their
working approach.
2.1 Mobile Ad Hoc Network (MANET)
Mobile Ad hoc Networks (MANET) are dynamic in nature where each node associates on Ad hoc or extemporaneous basis. MANETs appears in mid-1990 and
different research studies conducted on MANETs to improve its performance.
116
M. A. Hassan et al.
Routing between nodes does not depend on fixed infrastructure (base station) while
every node can directly send and receive data packets and can act as a relay node in
some specific situation [12].
2.2 Vehicular Ad Hoc Network (VANET)
Mobile Ad hoc Networks features are implemented in to vehicles to avoid rush
hours traffic and communicate with each other and road side sensors created another
wireless domain Vehicular Ad hoc network (VANET). VANET can support dense
network and semi-organized where every vehicle can directly communicate with
each other and send and receive data packets [13].
2.3 Flying Ad Hoc Network (FANET)
Flying Ad hoc network (FANET) is another class of MANET and VANET [15, 29]
as shown in Fig. 4 which adopt some basic features from its predecessors. Operated
in air making network on Ad hoc bases (temporary). Huge amount of growth has
been noticed in Unmanned Aerial Vehicle (UAV) industry from couple of decades.
According to report it will be double from $45 billion market value [14]. UAVs has
multiple applications and mostly used in military [16], commercial [17] and civil
applications [18] like disaster management [19], smart city [20] search and rescue
operations [21], fire checking and estimation [22], boarder security [23], Monitoring
of Traffic [24], wildlife and agricultural [25], object detection and tracking [26],
Fig. 4 Ad hoc networks
hierarchy
Impact of Routing Techniques and Mobility …
117
real time data collection [27], path planning and navigation [28]. Small size UAV
has extra importance on large size UAVs because of easy deployment, ease of hardware and software deployment and most importantly it is not threatening human
lives or property it can operate much low altitudes due to its small size. Mobile
Ad hoc networks (MANET) and VANET are limited to 2D direction movements
(x, y-axis) while FANETs can operate in three dimension (3D) (x, y, z-axis) with
speed of (30–460 km/h) [32] comparatively much faster than MANET (6 km/h) and
VANET (100 km/h in urban and 50 km/h in rural) [30, 31]. Speed of UAVs causing
rapid topology changes and network facing communication issues (disconnections)
between UAVs and Base station. Single, multiple and multiple group applications
are used in FANETs.
2.4 Single, Multiple and Multiple-group UAVs Application
Network
In single Unmanned Aerial Network applications, ground base nodes communicate with UAV network. Star shaped topology has been created by UAV where one
UAV acts as leader (backbone) of the all connected UAV. Data has been passed
through backbone and reached to the desired Base Station (BS). Moreover, in single
UAV application peer-to-peer communications are supported for rapid exchanging
of information as shown in Fig. 5. However, the Single Unmanned Aerial Network
framework allows them to challenge peer-to-peer communication e.g. increasing the
transmission range causing high interruptions and End-to-End Delay problems and
harvest network resources solution to these problems network should be equipped
with Omnidirectional antennas [33].
Multiple UAV application covers more geographical area than single UAV
network. In multiple UAV application more the one sub network has been created
among UAV. All data has been passed through backbone UAV to reached to its desired
destination as shown in Fig. 5. In multi-UAV strong communication link is needed
between UAV and BS for better utilization of network resources and special type of
hardware is needed (Fig. 6).
Multi-layer is another type of UAV application in which more than multiple group
can communicate with each other and share data without any centralized mechanism.
Figure 7 represents multi-layer network where each UAVs can communicate via
backbone UAV.
2.5 Classification of UAVs
UAV is classified in to two types based on the wings Rotary Wings and Fixed Wing.
Rotary wings are capable of y-axis take-off and landing and can be used in the
118
M. A. Hassan et al.
Fig. 5 Single UAV ad hoc network
static point missions while fixed wing can stay in air for longer time and has good
aerodynamic design as shown in Fig. 8.
2.6 Mobility Models
Mobility models also have an impact on the performance of FANET which predict
movements of UAVs, their speed and direction. Selecting suitable mobility models
for obtaining maximum utilization of network resources is critical task in Ad hoc
networks. Table 1 presents overall mobility model which has been deployed and
tested in Ad hoc network [34, 35].
Impact of Routing Techniques and Mobility …
119
Fig. 6 Multiple UAV ad hoc network
Fig. 7 Multiple-group UAV ad hoc network
2.6.1
Random Way Point
In this mobility model, pause time, maximum and minimum speed is set before
deploying a network for surveillance. Each node is directed towards random position
in a given space and moves for certain period of time. After covering some distance,
it then pauses and change velocity and acceleration [36–38].
120
M. A. Hassan et al.
Fig. 8 Fixed and rotary wings UAVs
2.6.2
Random Direction
This mobility model address density wave problem faced by Random Way Point
because of non-equal distribution of neighboring nodes. Center of simulation is
mostly affected by non-equal distribution of nodes. It also has pause time and random
direction selection mechanism. Only difference is uniform distribution of nodes in
second phase [39, 40].
2.6.3
Gauss-Markov
Sudden displacement of nodes is avoided by this mobility model. Tuning parameter
is responsible of different levels of nodes random direction. At first all nodes have
predefined speed and direction and later on future direction is decided based on
previous patterns (memory based). Sharp turn and sharp motion are also address for
maximum output [41, 41, 43].
2.6.4
Spring Mobility Model
Network is divided in to equal number of clusters where one node is selected as
Cluster Head (CH). In whole simulation CH is stationary and all other member
nodes are moving in to predefined pattern. Member nodes are also allowed to change
their group for specific amount of time called cycle time and then automatically
return to its base group [44, 45].
2.6.5
Distributed Pheromone Repel
In this mobility model all nodes broadcast its movements with its neighboring pals
to create pheromone mapping. It’s a grid type segment where each segment has its
own time stamp ensuring last was completely covered by each node [46].
Impact of Routing Techniques and Mobility …
121
Table 1 Comparison of different mobility models
Mobility Model
Advantages
Disadvantages
Applications
Random way point Easy to Implement and
can handle random
movements of nodes in
a network
No collision detection or Utilized in various
collision avoidance
application of ad hoc
mechanism and irregular networks
spatial distribution of
nodes
Random direction
Easy to deploy and can
tackle random
movements of nodes in
a network
No collision avoidance
Mostly deployed in
mechanism and does not MANET applications
have aerodynamics and
mechanical control
Gauss-Markov
(GM (3D))
Address sudden
movement changes and
suitable for rapid
changing topology
network
There is no collision
Avoidance feature in a
network and limited to
sharp turning of the
nodes in a network
Mostly used in
military application,
search and rescue
operations
Spring mobility
model (SMM)
Nodes in a network has
very smooth
acceleration
Address collision
problem in previous
mobility models
Delivery of goods
Distributed
pheromone repel
(DPR)
Smooth turning and
acceleration mechanism
and efficient network
courage
Does not take nodes
connectivity in to
account and fixed radius
turn of nodes
Search and rescue
operations and
patrolling in boarder
area
Smooth-turn (ST)
Address aerodynamic
constrains and efficient
in frequent topology
changing environment
Nodes can collide with
each other during
surveillance
Patrolling and for
military purpose
Semi-random
circular movement
(SRCM)
Very efficient trajectory
motion and overcome
collision among nodes
in a network
Fixed radius turns of
nodes in a network
Used in network
coverage in distant
area, fire and traffic
monitoring
Paparazzi mobility
(PPRZM)
Less sharp tern of nodes Nodes can collide with
in a network
each other
Random walk
Allowing nodes to move Sudden changing
mostly used in all ad
in random direction in
movements of nodes in a hoc network
fixed interval of time
network makes network applications
suffer with energy
problem
Manhattan grid
mobility model
Nodes selects random
direction in a given
space
Not efficient in less
dense network
Boundless
simulation area
It supports three
dimensional nodes
movements
Small geographical Area Mostly used in search
makes nodes
and rescue and
non-uniform division of monitoring
clusters
Environment sensing
and agriculture
monitoring
Most used in urban
vehicle movement
estimation
applications
(continued)
122
M. A. Hassan et al.
Table 1 (continued)
Mobility Model
Advantages
Enhanced
Gauss-Markov
Each nodes direction is Pausing of nodes makes
computed accurately
problem in a network
and added with collision
avoidance mechanism
Disadvantages
Applications
Applied mostly in
FANETs applications
Smooth turn
Allow to track nodes
Not efficient in dense
trajectory in the network network
Mainly utilized for
the MANET and
FANET applications
Three-way random Random path selection
using probability
matrices
Randomness of nodes
effects other parameters
of network
Mostly used in flying
IoT applications
Semi-random
circular movement
Nodes in a network opt
random direction in
given time to reach the
center of radius
Limits nodes to unique
fixed radius
Mostly used in
VANET and FANET
combination
Nomadic
community
Conceptual model e.g.
nodes Movements and
nodes localization
Inflexible movements of
nodes in a network
Agriculture and
military applications
Purse mobility
model
Grouping of nodes to
find particular target
Restricted behavior of
nodes
Tracking in vehicles
in dense V2V
network
Particle swarm
mobility model
Position of nodes are
calculated based on
reference points
Misleading of future
prediction of mobility
Civil and military
applications
Spatiotemporally
correlated group
mobility model
Spatial correction and
temporal trajectory of
nodes
Adopts gauss Markov
limitation which does
not provide theoretical
prove of OLS
Mostly used in
research and rescue
operations
Multi-Tier mobility Multiple nodes
model
movements are
supported
Collision of nodes
Mostly used in search
and rescue operation
Exponential
correlated random
Correlated group
Movements of nodes
Facing sharp turn
problems
Mostly utilized for
FANET applications
Distributed
pheromone repel
Ant colony-based
pheromone and
localization of nodes in
a network embedded
Unbalanced nodes
distribution
Delivery of goods
Self-deployable
point coverage
Connectivity is
maintained among
nodes on regular bases
Not effective in dense
network
Boarder surveillance
Impact of Routing Techniques and Mobility …
2.6.6
123
Smooth-Turn
It is specially designed for fast mobility and rapid changing topology network e.g.
FANET. Regular trajectory is generated by tends of nodes in a network. Due to its
simple design and accurate to analyze routing protocols of ad hoc networks it takes
advantage over other mobility model [47–49].
2.6.7
Semi-Random Circular Movement
Nodes in a network has virtual fixed radius for a specific period of time. After
successful round turn it then sets another virtual fixed radius and moves in that
radius with a fixed center [50].
2.6.8
Paparazzi Mobility
Nodes are working on path-based mechanism which in return opt five basic mobility
movements in a network and in return makes (i) eight type shape around target (ii)
Oval type shape (iii) stay-at a point (iv) scan (v) waypoint around the target [51].
Mobility Model is key parameter in FANET. Random Way Point Mobility model
is mostly utilized FANETs applications. In Table 1 all mobility models utilized in
FANETs are presented with its advantages, disadvantages and applications.
2.7 Routing Techniques
Different types of routing techniques have been adopted to accommodate various
types of constrains in Flying ad hoc networks. Similarly, mobility models are adopted
to overcome packet losses and delay problems. Each and every routing technique is
adopted when specific event is triggered.
2.7.1
Store Carry and Forward Routing Mechanism
SCF routing mechanism is adopted for those networks which suffers from intermittent
connectivity issues. Nodes buffers are used to store the data, and whenever these
nodes meet with other nodes in a network it then instantly forwarded the duplicated
data towards destination node [52].
124
2.7.2
M. A. Hassan et al.
Greedy Forwarding (GF)
This technique is more suitable for the dense network and reduce in between relay
nodes. Data packets can be traversed in a single routing communication path. Position
information of nodes are used to forward data towards the desired node [53].
2.7.3
Prediction Routing Mechanism (PR)
Highly mobile network use prediction-based routing mechanism to predict future
direction and speed of the nodes in a network. Robust and reliable communication
is possible depending upon correct predication [54].
2.7.4
Discovery Progress (DP)
This technique is used for highly mobile network in which discovery of nodes are
triggered when destination node is not known. Route Request (RRQ) is responsible
for the connection and node searching in a network [55].
2.7.5
Clusters (CL)
This mechanism is mostly utilized in the highly mobile and dense network. Where
network is divided in to zones or cluster depend on density of the network. After
creation of zones one nodes is elected as CH and responsible for inter and intra
communication and forwarding of data packets [56].
2.7.6
Link State (LST)
In link state each and every node in a network share its link information to its adjacent
neighbor node. Optimal path is utilized because of the link information sharing among
nodes. When node get disconnected it then immediately inform its neighbor node
[57].
2.7.7
Hierarchal (HR)
Network is divided in to upper or higher and lower levels to form a tree shape structure.
Each level has its own controlling node which is responsible for the connection and
communication arrangements with higher levels [58].
Impact of Routing Techniques and Mobility …
2.7.8
125
Mobility Information (MI)
All nodes movements, speed and velocities are predicted and help to select best
optimal relay node for sharing of data towards destination. Due to high speed of
nodes in a network required rapid exchanging of Hello packet for link discovery
[59].
2.7.9
Energy Efficient (EE)
This type of routing technique is adopted to control energy harvesting and ensure
maximum surveillance time of nodes in a network. Energy balance stabilizer is used
to maintain balance energy level of each node [60].
2.7.10
Static (STA)
In static routing mechanism predefined paths are set for monitoring. Nodes in a
network is enabled with topology information of whole network for optimal path
finding towards destination [60].
2.7.11
Secure (SC)
Data security is most in any network. In secure routing techniques network is more
secure and capable of detecting malicious node in a network [61].
2.7.12
Broadcast (BR)
Rapid changing topology network face regular disconnections. In broadcast routing
technique Hello packet is regularly spread throughout network to ensure successful
data transmission [62].
Different routing protocols are designed on these techniques. In Table 2 all routing
techniques are analyzed and presented with its advantages and disadvantages.
3 Conclusion and Future Direction
Flying Ad hoc Network (FANET) is considered as sub-class of MANET and VANET.
From the last decades we have noticed increase demand of FANET applications in
every sector. In COVID situation it is mostly utilized for delivery, monitoring and not
limited to these only. Routing and mobility models are considered as key component
to ensure right functionality and connected network of nodes. Many mobility models
126
M. A. Hassan et al.
Table2 Comparison of routing techniques
Technique
Advantage
Disadvantage
Store carry and forward
Intact network formation
Due to store mechanism it
produces more delay in the
network
Greedy forwarding (GF)
Minimum number of hops and
more important a smaller
number of transmission delay
Does not provide best
optimization at local optima
Prediction routing mechanism
(PR)
Nodes are more tightly
connected
Fails in less dense network
Discovery progress (DP)
Complete knowledge about
paths
Network face overhead
problem
Clusters (CL)
Divide Whole network in to
clusters
Poorly dense network is not
supported
Link state (LST)
Fully connected network
Large number of overhead
problems in a network
Hierarchal (HR)
Facilitate nodes with complete Failed in highly mobile
routing path towards destination scenarios
Mobility information (MI)
Network is fully connected
Fragmentation is not fully
supported
Energy efficient (EE)
Monitor power consumption
among connected nodes
Whole network connectivity
is not supported
Static (STA)
Packet delivery is more
improved
Rapid topology changes are
not supported
Secure (SC)
Host node deliver data to
destination
Complex mathematical
modulation is required
Broadcast (BR)
Granted data delivery
Network face large number
of Overheadproblems
and routing protocols are created for FANETs and other Ad hoc network each has
its own advantages and disadvantages and selecting of mobility model and routing
technique is depend on its utilization. In future we will analyze all routing protocols.
References
1. D. Mills, H. Braun, The NSFNET backbone network. ACM SIGCOMM Comput. Commun.
Rev. 17(5), 191–196 (1987)
2. P. Johansson, M. Kazantzidis, R. Kapoor, M. Gerla, Bluetooth: an enabler for personal area
networking. IEEE Netw 15(5), 28–37 (2001)
3. H. Lin, M. Choi, Mining web usage within a local area network. Int. J. Eng. Technol. 2(5),
435–441 (2010)
4. J. Huang, Cross layer link adaptation scheme in wireless local area network. J. Comput. Applic.
29(2), 518–520 (2009)
Impact of Routing Techniques and Mobility …
127
5. C. Dowin, Effect of network hierarchy in a typical campus area network (CAN) of a university.
Scient. Res. Essays 7(6) 2012
6. I. Song, Differences between seniors living in the metropolitan area and those in the nonmetropolitan area on the relationship between productive aging and depression—social network
as a moderator. J. Commun. Welfare (2013)
7. C. Dowd, NHS wide area network. Health Inform. 1(3), 83–83 (1995)
8. J. Nikolaidis, Practical storage area network [book review}. IEEE Netw 17(4), 4–4 (2003)
9. R. Horst, TNet: a reliable system area network. IEEE Micro 15(1), 37–45 (1995)
10. M. Weib, Passive wireless local area network radar network using compressive sensing
technique. IET Radar Sonar Navig. 9(1), 84–91 (2015)
11. H. Evers, Traditional trading networks of Southeast Asia. Archipel 35(1), 89–100 (1988)
12. D. Ergenç, L. Eksert, E. Onur, Dependability-based clustering in mobile ad hoc networks. Ad
Hoc Netw. 93, 01926 (2019)
13. J. Sun, Y. Fang, Defense against misbehavior in anonymous vehicular ad hoc networks. Ad
Hoc Netw. 7(8), 1515–1525 (2009)
14. H. Shakhatreh et al., Unmanned aerial vehicles (UAVs): a survey on civil applications and key
research challenges. IEEE Access 7, 48572–48634 (2019)
15. H. Tareque, M. Hossain, M. Atiquzzaman, On the routing in flying ad hoc networks, in 2015
federated conference on computer science and information systems (FedCSIS), (Lodz, Poland,
2015), pp. 1–9
16. H. Zhao, H. Wang, W. Wu, J. Wei, Deployment algorithms for UAV airborne networks toward
on-demand coverage. IEEE J. Sel. Areas Commun. 36(9), 2015–2031 (2018)
17. K. Sai Kumar, M. Sohail, P. Mahesh, U. Nelakuditi, Crowd monitoring and payload delivery
drone using quadcopter based UAV system, in 2018 international conference on smart systems
and inventive technology (ICSSIT), (Tirunelveli, India, 2018), pp. 22–25
18. S. Hayat, E. Yanmaz, R. Muzaffar, Survey on unmanned aerial vehicle networks for civil
applications: a communications viewpoint. IEEE Commun. Surv. Tutorials 18(4), pp. 2624–
2661 (2016). Available: https://doi.org/10.1109/comst.2016.2560343
19. I. Maza, F. Caballero, J. Capitán, J. Martínez-de-Dios, A. Ollero, Experimental results in multiUAV coordination for disaster management and civil security applications. J. Intell. Rob. Syst.
61(1–4), 563–585 (2010)
20. M. Khan, B. Alvi, A. Safi, I. Khan, Drones for good in smart cities: a review, in 2018 international conference on electrical, electronics, computers, communication, mechanical and
computing (EECCMC), (Tamil Nadu, India, 2018), pp. 1–7
21. M. Półka, S. Ptak, Ł Kuziora, The use of UAV’s for search and rescue operations. Procedia
Eng. 192, 748–752 (2017)
22. L. Merino, F. Caballero, J. Martínez-de-Dios, I. Maza, A. Ollero, An unmanned aircraft system
for automatic forest fire monitoring and measurement. J. Intell. Rob. Syst. 65(1–4), 533–548
(2011)
23. S. Kim, G. Lim, Drone-aided border surveillance with an electrification line battery charging
system. J. Intell. Rob. Syst. 92(3–4), 657–670 (2018)
24. M. Elloumi, R. Dhaou, B. Escrig, H. Idoudi, L. Saidane, Monitoring road traffic with a UAVbased system, in 2018 IEEE wireless communications and networking conference (WCNC),
(Barcelona, Spain 2018), pp. 1–6
25. H. Xiang, L. Tian, Development of a low-cost agricultural remote sensing system based on an
autonomous unmanned aerial vehicle (UAV). Biosys. Eng. 108(2), 174–190 (2011)
26. S. Zhu, D. Wang, C. Low, Ground target tracking using UAV with input constraints. J. Intell.
Rob. Syst. 69(1–4), 417–429 (2012)
27. N. Hossein Motlagh, T. Taleb, O. Arouk, Low-altitude unmanned aerial vehicles-based internet
of things services: comprehensive survey and future perspectives. IEEE Internet Things J. 3(6),
pp. 899–922 (2016)
28. C. Haohao, Q. Xiaohui, Y. Sen, Obstacle avoidance and path planning for quadrotor based on
differential evolution artificial bee colony algorithm. J. Phys. Conf. Ser. 1087, p. 022030 (2018)
128
M. A. Hassan et al.
29. M. Khan, I. Khan, A. Safi, I. Quershi, Dynamic routing in flying ad hoc networks using topology
based routing protocols. Special Issue Adv. Drone Commun. State-of-the-Art Architect. 2(3),
1–15 (2018)
30. I. Khan, I. Qureshi, M. Aziz, T. Cheema, S. Shah, Smart IoT control-based nature inspired
energy efficient routing protocol for flying ad hoc network (FANET). IEEE Access 8, 56371–
56378 (2020)
31. D.T. Senthil Murugan High mobility supports routing protocol (HMSRP) in mobile ad hoc
networks. Int. J. Eng. Technol. 7(24), p. 59 (2018)
32. X. Li, F. Deng, J. Yan, Mobility-assisted adaptive routing for intermittently connected FANETs.
IOP Conf. Ser. Mater. Sci. Eng. 715, 012028 (2020)
33. G. Kakamoukas, P. Sarigiannidis, A. Economides, FANETs in agriculture—a routing protocol
survey. Internet Things, p. 100183 (2020)
34. W. Zafar, B.M. Khan, Flying ad hoc networks: technological and social implications. IEEE
Technol. Soc. Mag. 35(2), 67–74 (2016)
35. J. Dentler, M. Rosalie, G. Danoy, P. Bouvry, S. Kannan, M.A. Olivares- Mendez, and H. Voos,
Collision avoidance effects on the mobility of a UAV swarm using chaotic ant colony with
model predictive control. J. Intell. Robot. Syst., pp. 1–17 (2018)
36. M.A. Hassan, S.I. Ullah, A. Salam, A.W. Ullah, M. Imad, F. Ullah, Energy efficient hierarchical
based fish eye state routing protocol for flying ad hoc networks. Indonesian J. Elect. Eng.
Comput. Sci. 21(1), 465–471 (2021)
37. M.A. Hassan, S.I. Ullah, I.U. Khan, S.B.H. Shah, A. Salam, A.W.U. Khan, Unmanned aerial
vehicles routing formation using fisheye state routing for flying ad hoc networks, in the 4th
international conference on future networks and distributed systems (ICFNDS), pp. 1–7 (2020)
38. J. Yoon, M. Liu, B. Noble, Random waypoint considered harmful, in Proceedings of the twentysecond annual joint conference of the IEEE computer and communications (INFOCOM’03),
vol. 2. IEEE, pp. 1312–1321 (2003)
39. G. Carofiglio, C.-F. Chiasserini, M. Garetto, E. Leonardi, Route stability in MANETs under
the random direction mobility model. IEEE Trans. Mob. Comput. 8(9), 1167–1179 (2009)
40. E. M. Royer, P. M. Melliar-Smith, L. E. Moser, An analysis of the optimum node density for ad
hoc mobile networks. in Proceedings of the IEEE international conference on communications
(ICC’01), vol. 3. IEEE, pp. 857–861 (2001)
41. VTolety, Load reduction in ad hoc networks using mobile servers, [117] S. Qazi, A. S. Siddiqui,
A.I. Wagan, UAV based real time video surveillance over 4G LTE, in Proceedings of the
International Conference on Open Source Systems & Technologies (ICOSST). IEEE, pp. 141–
145 (2015)
42. P. A. Regis, S. Bhunia, S.Sengupta, Implementation of 3D obstacle
compliantmobilitymodelsforUAVnetworksinns-3, in Proceedings of the workshop on ns-3.
ACM, pp. 124–131 (2016)
43. D.S. Vasiliev, A. Abilov, V.V. Khvorenkov, Peer selection algorithm in flying ad hoc networks,
in Proceedings of the international Siberian conference on control and communications
(SIBCON). IEEE, pp. 1–4 (2016)
44. F.N. Webber, R.E. Hiromoto, Assessing the communication issues involved in implementing
high-level behaviors in unmanned aerial vehicles, in Proc. IEEE Military Commun. Conf.
(MILCOM), (Washington, DC, USA, 2006), pp. 1–7
45. X. Hong, M. Gerla, G. Pei, C.-C. Chiang, Agroupmobility model for ad hoc wireless networks,
in Proc. 2nd ACM int. workshop on Modeling, analysis and simulation of wireless and mobile
systems, (Seattle, Washington, USA, 1999), pp. 53–60
46. E. Kuiper, S. Nadjm-Tehrani, Mobility models for UAV group reconnaissance applications, in
Proc. IEEE int. conf. on wireless and mobile commun. (ICWMC), (Bucharest, Romania, 2006),
pp. 33–33
47. Y. Wan, K. Namuduri, Y. Zhou, S. Fu, A smooth-turn mobility model for airborne networks.
IEEE Trans. Veh. Technol. 62(7), 3359–3370 (2013)
48. X. Xi, X. Cao, P. Yang, Z. Xiao, D.O. Wu, Efficient and fair network selection for integrated
cellular and drone-cell networks. IEEE Trans. Vehicular Technol. (2018)
Impact of Routing Techniques and Mobility …
129
49. S. Li, Y. Wan, S. Fu, M. Liu, H.F. Wu, Design and implementation of a remote UAV-based
mobile health monitoring system,” in Proceedings of the nondestructive characterization and
monitoring of advanced materials, aerospace, and civil infrastructure, vol. 10169. International
Society for Optics and Photonics, p. 101690A (2017)
50. W. Wang, X. Guan, B. Wang, Y. Wang, A novel mobility model based on semi-random circular
movement in mobile ad hoc networks. Inf. Sci. 180(3), 399–413 (2010)
51. O. Bouachir, A. Abrassart, F. Garcia, N. Larrieu, A mobility model for UAV ad hoc network,
in Proceedings of the international conference on unmanned aircraft systems (ICUAS). IEEE,
pp. 383–388 (2014)
52. Q. Usman, O. Chughtai, N. Nawaz, Z. Kaleem, K. A. Khaliq, L.D. Nguyen, Lifetime improvement through suitable next hop nodes using forwarding angle in FANET, in 2020 4th international conference on recent advances in signal processing, telecommunications & computing
(SigTelCom). IEEE, pp. 50–55 (2020)
53. L.A.L.F. da Costa, R. Kunst, E.P. de Freitas, Q-FANET: Improved Q-learning based routing
protocol for FANETs. Comput. Netw. 198, 108379 (2021)
54. Q. Sang, W. Honghai, L. Xing, P. Xie, Review and comparison of emerging routing protocols
in flying ad hoc networks. Symmetry 12, 6–971 (2020)
55. A. Srivastava, J. Prakash, Future FANET with application and enabling techniques: anatomization and sustainability issues. Comput. Sci. Rev. 39, 100359 (2021)
56. M.M.S. Ibrahim, P. Shanmugaraja, Optimized link state routing protocol performance in flying
ad hoc networks for various data rates of un manned aerial network. Mater. Today Proceed. 37,
3561–3568
57. M. Namdev, S. Goyal, R. Agarwal, An optimized communication scheme for energy efficient
and secure flying ad hoc network (FANET). Wireless Personal Commun, 1–22 (2021)
58. M.A. Khan, A. Safi, I.M. Qureshi, I.U. Khan, Flying ad hoc networks (FANETs): a review of
communication architectures, and routing protocols, in 2017 First international conference on
latest trends in electrical engineering and computing technologies (INTELLECT). IEEE, pp
1–9 (2017)
59. H. Yang, Z. Liu, An optimization routing protocol for FANETs. EURASIP J. Wirel. Commun.
Netw. 2019(1), 1–8 (2019)
60. A. Nadeem, T. Alghamdi, A. Yawar, A. Mehmood, M.S. Siddiqui, A review and classification
of flying Ad Hoc network (FANET) routing strategies. J. Basic Appl. Scientific Res. 8 (3), 1–8
(2018)
61. A. AlKhatieb, E. Felemban, A. Naseer, Performance evaluation of ad hoc routing protocols
in (FANETs), in 2020 IEEE wireless communications and networking conference workshops
(WCNCW). IEEE, pp 1–6 (2020)
62. I.U. Khan, M.A. Hassan, M.D. Alshehri, M.A. Ikram, H.J. Alyamani, R. Alturki, V.T. Hoang,
Monitoring system-based flying IoT in public health and sports using ant-enabled energy-aware
routing. J. Healthcare Eng. (2021)
Analysis of Vulnerabilities
in Cybersecurity in Unmanned Air
Vehicles
Mohammad Ammar Mehdi, Syeda Zillay Nain Zukhraf, and Hafsa Maryam
Abstract This paper studies the logical and exchange writing on network protection
for automated elevated vehicles (UAV), focusing on genuine and reenacted assaults,
and the ramifications for little UAVs. The audit is spurred by the expanding utilization
of little UAVs for investigating basic foundations such as the electric utility transmission and appropriation lattice, which could be an objective for illegal intimidation.
The paper presents a detailed context about different type of threats and attacks. The
paper additionally isolates and examines the discoveries by huge or little UAVs, over
or under 25 kg, yet thinks on little UAVs. The study reasons that UAV-related examination to counter network safety dangers centers around GPS Jamming furthermore,
Spoofing, because cybercrime can occur anytime on these UAVs and results will be
endless. It is impossible to deal with cyber-attacks without knowing the threats and
the nature of attacks. In this paper, we summarize the major types of threats/attacks
and the parameters that are affected by these threats/attacks.
Keywords UAVs · Attacks · Threats
1 Introduction
In a world of consistent and fast mechanical modification, limiting weaknesses may
be a unremitting race against one’s enemies—their innovation, their gadgets, their
thoughts, their ways of activity, their methods, and their double-dealing of innovation
patterns to accomplish their political objectives. UAV became a lot of traditional, all
M. A. Mehdi
Department of Electrical Engineering, The Islamia University of Bahawalpur, Bahawalpur,
Pakistan
S. Z. N. Zukhraf (B) · H. Maryam
Department of Electrical and Computer Engineering, KIOS Research and Innovation Centre of
Excellence, University of Cyprus, Nicosia, Cyprus
e-mail: zukhraf.syeda-zillay@ucy.ac.cy
H. Maryam
e-mail: maryam.hafsa@ucy.ac.cy
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_8
131
132
M. A. Mehdi et al.
Fig. 1 Exemplary illustration of common cyber-attacks
the more promptly accessible, and a lot of advanced, supporting new capacities like
dilated info assortment and freelance conduct. As a result, UAV square measure
reshaping the net protection world in two keyways in which. Right off the bat,
UAV gift another style of basic network protection target. Basic law demand or
info assortment missions utilizing UAV may be sabotaged by cyber-attacks on these
stages. Also, UAS within the possession of foes may introduce novel roads for cyberattacks, with UAS themselves filling in as “digital weapons” expected to convey
pernicious substance or motor effects [1]. Figure 1 shows the scenario of random
attacks.
The dependence and utilization of robots is regularly ascending in numerous areas.
This is often owing to the robots’ capability to supply a live-transfer, current video and
movie catch, aboard the capability to fly and ship merchandise [2]. Consequently,
in way over 10,000 robots are going to be purposeful for business use within the
subsequent 5 years. This is often preponderantly owing to their edges over business
helicopters with regards to expenses and finances [3]. Additionally, the mechanical
headway empowers straightforward controls through advanced cells to fly very little
robots as opposition utilizing distant regulators. Indeed, the employment of robots
is not restricted to business and individual points.
Robots square measure being used by law implementation and boundary management intelligence teams. Within the event of harmful events, search and salvage
teams utilize them to accumulate knowledge or to drop elementary provides. Be that
because it might, drones don’t seem to be being used entirely by “heroes”; “troublemakers” square measure utilizing robots to accomplish their harmful goals. Being not
tough to regulate, robots are often used to perform numerous assaults. Then again,
drones uncover security weaknesses that build them inclined to commandeering.
Analysis of Vulnerabilities in Cybersecurity …
133
During this paper, we tend to survey the assaults from/to drones, aboard their current
countermeasures.
The security of non-military personnel drones was audited in [4]. Likewise, unique
security assaults on rambles were cleft in [5–9]. Robots’ location techniques were
evaluated in [10, 11]. Yet, a elementary restriction of the past work is that the
absence of a way reaching examination of the robots security weaknesses and also the
assault life cycle. Besides, simply one a part of robots’ security dangers was caredfor assaults on rambles. The present countermeasures ought to be investigated, and
new procedures ought to be projected to beat the restrictions of the present security
arrangements.
2 Motivation
Due to the internet connectedness, our contemporary society is being encircled by
gadgets and services. This all is possible because of the evolution in technologies.
These progress in technologies create our life vogue straightforward, services on one
click away and most significantly it narrowed the communication gap. It’s creating
our lives relaxed on one facet however on the opposite hand it results in several
security flaws if we have a tendency to don’t take security precautions. Based on
factual and commerce literature, the demand for the supply of remote-controlled
Aerial Vehicles (UAVs) is predicted to extend each year by a awfully huge margin
by 2024 [12]. Little UAVs that weight but twenty-five kg grams, expressed by the
Drone Energy Alliance in United States, are the theatre of preference owing to the
affordable costs and also the ease to use them. Our modern urban areas, technically
called Smart Cities, are clustered by UAVs for various goal to make the lives of human
being easy. Drones are manageable and very advantageous if used properly. With all
of these advantages there are some technical objections and major issues for UAVs.
One amongst the foremost issues is Cyber security [13]. Drones are used personal,
commercial, and industrial purposes and they are used for military functions too.
Once put-upon then it can direct to the incursion of the privacy. Just In case of armed
drones used for Military purposed, if they face any kind of breach then the results
would be unfortunate. UAVs have wireless communication medium installed in them
for communication purpose. The attackers hack a drone through that communication
medium [14]. The study focuses on both renowned and speculative hazards from
events for small UAVs. It additionally contains material on cyberattacks on any size
UAV as such risks might apply to small UAVs.
134
M. A. Mehdi et al.
3 Cybersecurity Threats
A threat is an insecurity which may manipulate an accountability to invade security
and cause damage. These may be done “willingly” (that is hacking, criminal establishment) and “Unwillingly” (that is technical issue, natural disaster). In this section
Wireless Network UAVs threats are discussed based on STRIDE modeling [15, 16].
Below is the Fig. 2 Explaining the Threats mentioned in detail in a pictorial form.
3.1 Spoofing
Spoofing is the act of impersonating a valid source or process. This frequent type of
attack has previously been used in real-world power grids. For example, the Ukraine
power grid was hacked. It was launched in a such way that the hacker sent spam email
to workers in order to gain access to the supervisory control and data acquisition
system [17–19]. Many flaws were discovered during an testing of the design and
flight controllers of UAV prototypes with several rotors. These are the issues that are
related with both the communication links feeding information to/from a drone via
serial port links, particularly due to its poor communication characteristics, that is
not secured in many situations [19, 20]. The data can be readily collected, changed,
or inserted via GPS spoofing. This data connection weakness allows for blocking
and spoofing, allowing intruders full control of the UAV as shown in [21, 22].
Fig. 2 Explanation of the
word STRIDE
Analysis of Vulnerabilities in Cybersecurity …
135
3.2 Tampering
Tampering is defined as the illegal modification or elimination of information or a
process. This is a false attack in which information is inserted by the attackers and they
take cash-in of the weaknesses in hardware and communication links [17]. Tampering
is the act of interfering with the authenticity of a system under cyberattack by altering
that in another way. Tampering might arise in the context of UAV-related cybersecurity, when UAVs are used as a cyber weaponry, if a UAV is used to transfer spyware
to a targeted device to acquire an unprotected wireless network. Such malware has
the ability to defect high-value gear, such as industrial or power station equipment, as
well as assault slightly elevated targets including irrigation transmission and distribution networks grids [23]. However, this method was discovered to be unsafe. It
enables intruders to construct a reverse-shell TCP payload and implant it into the
UAV’s storage, where it would surreptitiously install malware on the ground-based
computers [1].
3.3 Repudiation
The perpetrators do not take the responsibility for their attack. This danger is the
least vital within the context of UAS protection. Internal abuse of system controls
is one type of repudiation. A drone controller, for instance, can argue that he or
she did not intentionally damage a device by claiming the failure on a design fault
in the network connection. Another example, when UAS are cyber munitions, may
be to disassociate an attacker’s identification from the end result by disrupting at a
connectivity node that is solely feeble related to the location of damage or disruption
[1].
3.4 Information Disclosure
It refers to violations of the secrecy principle. An agent exposes information to
someone who does not have the required credentials to receive it in an informationdisclosure attack. Infiltrating a UAS detection information system to get access to
video, audio, or alternative information might be a data-disclosure hazard. An agent
may additionally reveal information and then use repudiation to deny responsibility
later.
136
M. A. Mehdi et al.
3.5 Dos
DOS stands for Denial of service. It is referred to as denying the ease of access of
a resource that is required for the attacked system to work in a proper manner. An
example of DoS is when UAV are aimed and could involve contaminating drone
control software to make the devices unresponsive to user inputs.
3.6 Elevation of Privilege
This is a type of threat in which is possible by violating the principle of authorization
to hold out an action that one is not always allowed to do. An instance of authorization
of privilege is whilst UAV targets and could contain hijacking of a drone via posing
as a legitimate controller. While UAVs are used as a cyber weapon, they can be used
to deliver records, code, or different indicators to debilitate or alter the behavior of
the system underneath attack [1].
4 Attacks
4.1 Spoofing Attack
GPS plays a vital role in Unnamed Aerial Vehicle (UAV). Also, GPS spoofing is a
primary security attack on UAV in which attacker intended to drive a fully equipped
UAV along a trajectory towards any malicious target destination via real time GPs
without triggering the detector. Iran captured a U.S surveillance RQ-170 drone with
only minor damage, happened in December 2011 [24]. This incident left a huge
impact of GPS spoofing attack in terms of capture and control of the UAV. An Iranian
engineer claimed in an interview that GPS spoofing attack has been used on drones
GPS by using false coordinates and jammed the communication signals during this
operation for the force landing of UAV. To validate the Iranian’s claim, University of
Texas researchers conducted research in 2012 to demonstrate a short-term control on
UAV’s [22, 25]. In 2013 and 2014, [26] conducted the three types of cyber-attacks
for UAV’s are:
1.
2.
3.
Actuators compromised
Sensor compromised
Actuators and sensors compromised.
State estimation error of the UAV has been analysed and derived without being
detected. Also, sensor compromised based scenarios has been considered as a GPS
spoofing attack. Later [26] performed off-line computation and obtained all the
feasible injected values. In 2015, they performed real-time GPs spoofing attack in
Analysis of Vulnerabilities in Cybersecurity …
137
Fig. 3 Spoofing attack illustration
which sequential probability ratio detector has been discussed [27, 28]. However,
authors did not provide the detailed explanation of the UAV’s driving from source
to malicious destination [29–31] conducted the research in which they injected the
falsified GPS measurement to the navigation. Figure 3 gives an over view of the
attack.
4.2 Man in the Middle Attack
It is an attack in which attacker access the sensitive information without the user
consent and monitors the contract between two parties. In other words, attacker gain
the complete control over the communication path between two devices and redirect
all the communication. Attacker target’s is the original device on which all activities
are monitored between two devices by replacing the devices with the Wi-Fi router
at the receiver end. Rodday et al. [32] conducted a real-time experiment in which
MitM attack has been launched by injecting the control commands using packets.
To the best of our knowledge these authors has only conducted a real time research
on UAV’s. Figure 4 gives the overview of the discussed scenerio.
138
M. A. Mehdi et al.
Fig. 4 Man-in-middle attack illustration
4.3 DoS Attack
It is characterized to prevent the communication between two network devices from
using desired network resources. Gudla et al. [33] has been performed a cyber-attack
on AR. Drones by using an intrusion tool for the communication of UAV. The author
used the de-authentication key as an interruption and crashed the UAV. It is the type of
spoofing in which the attacker attacks the authorized user and broke the transmitter,
controller receives the warning regarding the link interruption and the UAV stops
at the present location. DoS attack can be performed using GPS tracking for the
investigation of data using log files. Vasconcelos et al. [34] performed attack on Ar.
Drone by analyzing the UAV mid-air ports and DoS assault strategies. Figure 5 shows
the illustration of the discussed attack.
4.4 Buffer Overflow Attack
It is just like a DoS attack and overloads the UAV memory with the computational capabilities of the device. Also, attacker block the network administrators
from accessing the resources on their devices. The attacker uploads different files by
submitting 1000 UAV queries concurrently. As a result, navigations system of UAVs
became stopped, collapsed, and crashed. Hooper et al. [35] conducted the real time
experiment in which they struck the parrot drone three times and evaluated the traffic
using Wireshark and analyzed the available ports with Nmap before triggering the
attack. Figure 6 shows the pictorial representation of the attack.
Analysis of Vulnerabilities in Cybersecurity …
139
Fig. 5 DoS attack illustration
Fig. 6 Buffer overflow attack illustration
4.5 Eaves Dropping Attack
It is based on confidentiality attack in which private data is accessed illegitimately.
This attack only become fruitful when receiver received an excellent Signal-to-Noise
Ratio (SNR) and the data is in encrypted form [36]. In this attack, attacker use the
passive attack; however, to enhance the performance of data interception active attack
can be combined. In passive attack of UAV’s attacker acts as a spy, difficult to locate
and simply listen all messages between two entities. Sometimes, In UAV’s, attacker
involves in a communication by listening and modifying a message and remain
140
M. A. Mehdi et al.
unnoticed to other. Sun et al. [37] proposed the route optimization and allocation of
joint resources to avoid eves droppers which prevent the leakage of information and
transmission power. Figure 7 shows the flow of attack.
The below Table 1 contains the analysis of all the factors affected by these attacks
and threats.
Fig. 7 Eaves dropping attack illustration
Table 1 Parameters for attacks & threats in UAVs
Attack
Threat
Attacks/Threats
Privacy
Integrity
Confidentiality
Authentication
Trust
Spoofing
No
No
No
Yes
No
Man-in-middle
Yes
Yes
Yes
No
No
DoS
No
No
No
Yes
No
Buffer overflow
No
No
No
Yes
No
Eaves dropping
Yes
No
Yes
No
No
Spoofing
No
No
No
Yes
No
Tempering
Yes
No
No
No
No
Repudiation
No
No
No
No
Yes
Information
disclosure
Yes
Yes
Yes
No
No
DoS
No
No
No
Yes
No
Elevation of
privileges
No
No
No
No
Yes
Analysis of Vulnerabilities in Cybersecurity …
141
5 Conclusion
The extraordinary growth for the usage of drones and UAVs caused a new aviation technology of self-sustaining aerial vehicles in both the civilian and military
domains, providing sever advantages including budget friendly, industrial, industrial, especially because of their independent, flexible, and easy-to-use nature, with
low fee and electricity intake. But their use brought about the increase in protection and privateness problems. Due to progressing world and emerging technologies
the threats of cybersecurity have also increased [38]. In this context, we described
the most common cybersecurity threats and attacks that make our system weak.
subsequently, due to the alarmingly growth inside the use of drones via terrorists, in
addition studies and experiments on a way to prevent and counter the UAV threats,
imposed by way of terrorists, can be finished, and carried out as a part of destiny
paintings. As the time pass by and modern technology excels the threats are also
arising day by day due to which we need to continue our research to defend our
systems and our privacy from emerging threats.
References
1. N. Shashok, Analysis of vulnerabilities in modern unmanned aircraft systems. Tuft University,
pp. 1–10 (2017)
2. K. Chan, U. Nirmal, W. Cheaw, AIP conference proceedings, vol. 2030. AIP Publishing.
Progress on drone technology and their applications: a comprehensive review, p. 020308 (2018)
3. Z. Liu, Z. Li, B. Liu, X. Fu, I. Raptis, K. Ren, Proceedings of the 2015 workshop on privacyaware mobile computing. ACM. Rise of mini-drones: applications and issues, pp. 7–12 (2015)
4. R. Altawy, A.M. Youssef, Security, privacy, and safety aspects of civilian drones: a survey.
ACM Trans. Cyber-Phys. Syst. 1(2), 7 (2017)
5. D. He, S. Chan, M. Guizani, Drone-assisted public safety networks: the security aspect. IEEE
Commun. Mag. 55(8), 218–223 (2017)
6. M. Yampolskiy, P. Horvath, X.D. Koutsoukos, Y. Xue, J. Sztipanovits, Proceedings of the 2nd
ACM international conference on High confidence networked systems. ACM. Taxonomy for
description of cross-domain attacks on cps, pp. 135–142 (2013)
7. H. Sedjelmaci, S.M. Senouci, Cyber security methods for aerial vehicle networks: taxonomy,
challenges and solution. J. Supercomput. 2018, 1–17 (2018)
8. T. Humphreys, Statement on the vulnerability of civil unmanned aerial vehicles and other
systems to civil gps spoofing. Univer. Texas Austin, 18 July 2012
9. D.P. Shepard, J.A. Bhatti, T.E. Humphreys, A.A. Fansler, Proceedings of the ION GNSS
Meeting, vol. 3. Evaluation of smart grid and civilian uav vulnerability to gps spoofing attacks,
pp. 3591–3605 (2012)
10. İ. Güvenç, O. Ozdemir, Y. Yapici, H. Mehrpouyan, D. Matolak, 2017 IEEE/AIAA 36th
digital avionics systems conference (DASC) IEEE. Detection, localization, and tracking of
unauthorized UAS and jammers, pp. 1–10 (2017)
11. R.L. Sturdivant, E.K. Chong, Systems engineering baseline concept of a multispectral drone
detection solution for airports. IEEE Access. 5, 7123–7138 (2017)
12. Navigant, Drones and robotics for utility transmission and distribution: unmanned aerial vehicle
and robotics solutions for utility T&D inspection and maintenance: global market analysis and
forecasts
142
M. A. Mehdi et al.
13. https://www.navigantresearch.com/research/dronesand-robotics-for-utility-transmission-anddistribution, 2015. Accessed 29 Mar 2017
14. K. Nonami, Prospect and recent research & development for civil use autonomous unmanned
aircraft as UAV and MAV. J. Syst. Des. Dyn. 1(2), 120–128 (2007)
15. V. Dey, V. Pudi, A. Chattopadhyay, Y. Elovici, Security vulnerabilities of unmanned aerial
vehicles and countermeasures: an experimental study, in 2018 31st international conference
on VLSI design and 2018 17th international conference on embedded systems (VLSID). IEEE,
pp. 398–403 (2018)
16. R. Khan, K. McLaughlin, D. Laverty, S. Sezer, STRIDE-based threat modeling for cyberphysical systems, in 2017 IEEE PES Innovative Smart Grid Technologies Conference Europe
(ISGT-Europe). IEEE, pp. 1–6 (2017)
17. C.L. Krishna, R.R. Murphy, A review on cybersecurity vulnerabilities for unmanned aerial
vehicles, in 2017 IEEE international symposium on safety, security and rescue robotics (SSRR).
IEEE, pp. 194–199 (2017)
18. D.U. Case, Analysis of the cyber attack on the Ukrainian power grid. Electricity Information
Sharing and Analysis Center (E-ISAC), p. 388 (2016)
19. B. Ly, R. Ly, Cybersecurity in unmanned aerial vehicles (UAVs). J. Cyber Security Technol.
5(2), 120–137 (2021)
20. Y. Zeng, R. Zhang, T.J. Lim, Wireless communications with unmanned aerial vehicles:
opportunities and challenges. IEEE Commun. Mag. 54(5), 36–42 (2016)
21. S.H. Seo, B.H. Lee, S.H. Im, G.I. Jee, Effect of spoofing on unmanned aerial vehicle using
counterfeited GPS signal. J. Position. Navigat. Timing 4(2), 57–65 (2015)
22. A.J. Kerns, D.P. Shepard, J.A. Bhatti, T.E. Humphreys, Unmanned aircraft capture and control
via GPS spoofing. J. Field Robot. 31(4), 617–636 (2014)
23. K.L. Best, J. Schmid, S. Tierney, J. Awan, N.M. Beyene, M.A. Holliday, R. Khan, K. Lee, How
to analyze the cyber threat from drones. RAND ARROYO CENTER SANTA MONICA CA
SANTA MONICA, United States (2020)
24. Wiki (2011). Irancu.s. rq-170 incident. https://en.wikipedia.org/wiki/Iran%E2%80%93U.S.
RQ-170 incident
25. D.P. Shepard, J.A. Bhatti, T.E. Humphreys, A.A. Fansler, Evaluation of smart grid and civilian
uav vulnerability to gps spoofing attacks, in Proc. ION GNSS, vol. 3 (2012)
26. C. Kwon, W. Liu, I. Hwang, Security analysis for cyber-physical systems against stealthy
deception attacks, in Proc. IEEE ACC, pp. 3344–3349 (2013)
27. C. Kwon, W. Liu, I. Hwang, Analysis and design of stealthy cyber-attacks on unmanned aerial
systems. J. Aerospace Inform. Syst 11(8), 525–539 (2014)
28. C. Kwon, S. Yantek, I. Hwang, Real-time safety assessment of unmanned aircraft systems
against stealthy cyber-attacks. J. Aerospace Inform. Syst. 13(1), 27–45 (2015)
29. H. Zhang, P. Cheng, L. Shi, J. Chen, Optimal denial-of-service attack scheduling with energy
constraint. Transact Automat Control 60(11), 13023–13028 (2015)
30. H. Zhang, P. Cheng, L. Shi, J. Chen, Optimal dos attack scheduling in wireless networked
control system. Transact. Control Syst. Technol. 24(3), 843–852 (2016)
31. Y. Qi, P. Cheng, L. Shi, J. Chen, Event-based attack against remote state estimation, in Proc.
IEEE CDC, pp. 6844–6849 (2015)
32. N.M. Rodday, R.D.O. Schmidt, A. Pras, Exploring security vulnerabilities of unmanned aerial
vehicles, in NOMS 2016–2016 IEEE/IFIP network operations and management symposium.
IEEE, pp. 993–994 (2016)
33. C. Gudla, M.S. Rana, A.H. Sung, Defense techniques against cyber-attacks on unmanned aerial
vehicles, in Proceedings of the international conference on embedded systems, cyber-physical
systems, and applications (ESCS), pp. 110–116 (2018)
34. G. Vasconcelos, G. Carrijo, R. Miani et al., The impact of DoS attacks on the AR. Drone
2.0, in 2016 XIII Latin American Robotics Symposium and IV Brazilian Robotics Symposium
(LARS/SBR). IEEE, pp. 127–132 (2016)
35. M. Hooper, Y. Tian, R. Zhou, et al., Securing commercial wifi-based UAVs from common
security attacks, in MILCOM 2016–2016 IEEE military communications conference. IEEE,
Nov, pp. 1213–1218 (2016)
Analysis of Vulnerabilities in Cybersecurity …
143
36. Q. Wu, W. Mei, R. Zhang, Safeguarding wireless network with UAVs: a physical layer security
perspective. IEEE Wirel. Commun. 26(5), 12–18 (2019)
37. X. Sun, D.W.K. Ng, Z. Ding, Y. Xu, Z. Zhong, Physical layer security in UAV systems:
challenges and opportunities. IEEE Wirel. Commun. 26(5), 40–47 (2019)
38. S.Z. Sajal, I. Jahan, K.E. Nygard, A survey on cyber security threats and challenges in modem
society, in 2019 IEEE international conference on electro information technology (EIT). IEEE.
Symantec internet security threat report published in February 2019. Source: https://symantec.
com/security-centre/threat-report. Last Access 1 March 2019. Graph
Silent Listening to Detect False Data
Injection Attack and Recognize the
Attacker in Smart Car Platooning
Sharmistha Majumder, Mrinal Kanti Deb Barma, Ashim Saha, and A. B. Roy
Abstract In this chapter, a False Data Injection attack is investigated for smart-car
platooning in real time. ARP spoofing based silent listening technique is utilized
to implement the attack detection and attacker recognition. Deep learning based
smart car model of automation level 5 is simulated in the Unity game engine. A
post-test only experiment set up of five experiment groups are prepared to launch
attack, detect attack and recognize the attacker in real-time. The objectives are to find
out optimal security threats in the whole system; and come up with an alternative
solution of game theoretic model. Authors successfully simulate deep-learning based
smart car environment as a test bed; and then launch False Data Injection attack in
the network. This work achieves 100% correctness of both the algorithms. It argues
more effectiveness of silent listening or ARP spoofing over game based solution for
attack detection with the help of two separate algorithms for attack detection and
attacker recognition.
Keywords Silent listening · ARP spoofing · IP forwarding · False data injection
attack · Vehicle platooning · Smart car test bed · Unity game engine
1 Introduction
Easily an attacker can target smart car; since it is mostly physically exposed [1, 2].
Car manufactures generally overlooks non-malicious conditions and focus more on
its functionality and comfort [1] compared to its security. But both experimentally
S. Majumder (B) · M. K. Deb Barma · A. Saha
National Institute of Technology Agartala, Agartala, India
e-mail: sharmistha.cse@nita.ac.in
A. Saha
e-mail: ashim.cse@nita.ac.in
A. B. Roy
Madanapalle Institute of Technology and Science, Madanapalle, India
e-mail: arijitbardhanroy@mits.ac.in
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_9
145
146
S. Majumder et al.
(under lab setup) and in real (on road testing), [3] showed us how even driver inputs
can be completely ignored by an attacker using False Data Injection (FDI). An
attacker first intercepts the actual speed update packet sent by BCM (Body Control
Module), modifies it and re-transmits the modified or false speed [3]. She can even
modify the actual status of a malfunctioned or a nonexistent airbag into a healthy
airbag [4]. Unfortunately, vehicle platooning is another worst reality [5]. Ever since
the history of cyber security, neither the intention of the attacker can be measured [6],
nor the attacking techniques have been found to be fixed. To make self-driving cars
secured under FDI attacks [7], cyber security analysts should first provide Optimized
System Security by detecting security breaches. From optimized security detection
point of view, probabilistic methods [6, 8] are theoretically claimed to be more sound
than reward processes dealt in like [9]. Further, it is worth mentioning that [6] argued
that probabilistic modeling is more appropriate than reward based modeling since
the latter is yet to be quantified. In this chapter, we are alternatively going to use
smart car and self-driving car in same sense.
Self-driving car gathers inputs from its environment through sensors; processes
these inputs in its computerized brain with the help of different complex logics like
artificial intelligence, soft computing, deep learning etc. and finally, sends output to
the environment through its actuators. Note that this environment is nothing but our
physical world. So, it will not be a fiction at all if a techie human being decides to
launch an attack with smart-cars as weapon. The motive behind such attack may be
anything like fame or money or personal grudge or mental problem or just fun. Boons
are thus becoming curse in seconds. To protect our physical surrounding, we have
to adapt protection against threats raised by such computerized nonhuman brains.
Here, in this work, we shall deal with self-driving car of driving automation level 5
[10].
In Sect. 2, we will see few existing attack detection algorithms. Most of these
existing works are dealing with game-based solutions. But before launching our
vehicles in physical world, this section suggests an experimental set-up, and an
easy attack-detection algorithm to measure the security loop holes. Precisely, this
chapter contributes to simulate a self-driving robotic car environment in gaming
mode. Simulated cars will send and receive sensor data in real time. This realtime self-driving car environment will be hacked using FDI and the attack-detection
algorithm proposed in the paper will be used to detect the attack accurately, under
cyber security expert’s supervision. By accuracy, here we mean to catch the actual
attacker and that too in real time scenario.
For our problem, we went through bunch of hacking tricks and then came up
with the solution of network monitoring with silent listening. ARP spoofing is a
type of attack in which a malicious attacker sends altered ARP (Address Resolution
Protocol) messages over same local area network (LAN). This results in the linking of
an attacker’s MAC (Media Access Control) address with the IP (Internet Protocol)
address of a authorized computer or server on the same network [11]. Once the
attacker’s MAC address is connected to an authentic IP address, the attacker will
begin receiving any data that is intended for that IP address [11]. ARP spoofing thus
enables to monitor, modify or even stop data in-transit [11]. Out of several ARP
Silent Listening to Detect False Data Injection Attack …
147
spoofing based attacks like denial-of-service (DoS), session hijacking, man-in-themiddle etc. The present work incorporates the working principle of silent listening
in attack-detection or finding out cyber security loop-holes.
Under our proposed experimental set-up, in real time our attack-detection algorithm reported 100% success to identify the actual attacker. The paper applied probabilistic method to verify, understand the methodology, results obtained. This type
of fascinating result encourages the authors to come up with systems for real-time
monitoring and attack-detection as future work.
2 Related Works
In [1], the authors meticulously presented numbers of available smart-car cybersecurity attacks. They [1] researched on the available attacks; and reached to the
conclusion that among lots of theoretical or simulation or abstract level research
works, only a handful FDI attacks are observed on real cars. Koscher et al. [3] and
Hoppe et al. [4] are the examples. Wolf et al. [12] told about security problems
dwelling in automotive bus system. The existence of a malicious attacker is a real
threat [12]. Interestingly, solutions for FDI attack exits [13]. From Sect. 1, it is already
clear that scope of the present work is limited only up to detecting security threats
in self-driving cars. So, we are going to concentrate more on attack detection, rather
than prevention. Wu and Wang [9] took idea of Nash Equilibrium to detect collaborative security for Internet of Things (IoT) systems. In large-scale wireless networks,
researchers came up with game-theory based attack detection algorithms where on
detecting an attack, the attacker node was punished by discontinued with network
connection [14]. From dozens of research papers on cyber security, we observed that
because of attacker’s psychology, remarkable attack detection and defense techniques
are theming on game theoretic approach specially. Han et al. [15]’s approach also
enjoyed game theoretic solution fused with clustering analysis and semi-supervised
learning for cloud computing. In another work [16], authors applied game theoretic
approach in form of jamming game to achieve equilibrium state of Nash Equilibrium
in cognitive radio network. Compared to these aforementioned reward processes [9,
14–16], we are interested more on event based processes. These papers [9, 14–16]
are showing calculation of the same Nash Equilibrium to analyze security threats
under different networks and different architectures. We also studied other game
theoretic approaches rather than Nash Equilibrium, like dynamic Bayesian signaling
game [17], where authors gave sufficient chance to the attacker to breach the system, which as per our understanding can be treated as an asset to learn the dynamic
behavior of an attacker. This behavior analysis is the most important aspect to measure cyber security threats residing within a system. We co-relate our solution with
it. Vigorous monitoring of the data in-transit can help the cyber-security analysts to
understand fishy and normal events. Present work aims at FDI attack detection and
recognizing the attacker in self-driving car network.
148
S. Majumder et al.
Development of automotive electronics was seen in late 1970s [12]. Immense
development took place in self-driving car due to DARPA challenges [18]. Then
Udacity paired with DiDi to gear up the development of self-driving cars through
Udacity-DiDi Self-Driving Challenge [18]. Multiple sensor nodes, deep neural networks all are coming to the scene now. Those were the days when we used to watch
self-driving cars in sci-fi movies only. Now it’s a reality. For that we hardly need to
have a gigantic looking so called robotic car! We can simulate our own deep neural
network based self-driving car [19]. With the advent of ImageNet [20], we stepped
into the era of Deep learning. While modeling a self-driving car as stated in [18], huge
number of image data collected from the self-driving car sensors are being trained
using Convolutional Neural Network or CNN [21]. We studied the efforts of [22] to
convert a robot into a self-driving car; of [23] to launch a self-driving version of a car
without changing much of its appearance; of [18] to efficiently fuse various sensor
data in real time. This study encouraged us to simulate self-driving car environment
in our computer node.
FDI attack is vulnerable against sensing, measuring, and monitoring system of
smart cars [5] to mislead a system. The author proved attacker’s capability of gaining
control over the positions and velocities of vehicle platoon, in one way or the other [5].
This work thus motivated us to work in FDI attack detection in smart-car networks.
However, our proposed work is different in three main aspects: first to create a
deep learning based real-time test bed for smart-car network, second to monitor real
time data in-transit without help of any costly network device, third to detect FDI
attack in our system in real time using the hacker’s technique. Analytic mind will
understand the cost saving approach in such a huge work.
We found that game theoretic approaches give an attacker freedom to launch
attacks. Though it has advantage for understanding attackers’ behaviour or patterns
of particular type of attack, but at the same time, we have to remember that the gain
from the attack can be a big loss in terms of money or data or even life. So, instead of
applying game models, we are going for network monitoring and silent listening. During more than one experiment session, we observed the behaviour of the network, data
and other parameters. We found that these behaviour analysis are very much important and we can train our system to understand this specific patterns and behaviours.
2.1 Summary of Contributions
Main contributions of the paper are summarized as follows:
• Creating a deep-learning based smart-car test-bed for real time security analysis.
• Applying hackers’ technique to monitor real time data in-transit.
• A real time attack detection as well as attacker recognition algorithm using silent
network monitoring or ARP spoofing.
• Calculating uniqueness of the test-bed with available professional smart-car systems.
Silent Listening to Detect False Data Injection Attack …
149
• Alternative approach rather than game-based solution.
• A cost-effective solution for big project.
2.2 Chapter Organization
The chapter is organized as follows. In Sect. 3, research study questions (RQ), sample,
procedure, instruments and data analysis technique are discussed in details. Results
are given in Sect. 4, followed by conclusion of our work summarized in Sect. 5.
3 Research Method
In this paper, the authors are trying to answer the following research study questions
(RQ).
1. Can we simulate real time, deep-learning based smart car network as our test
bed?
2. Can we use alternative approach to identify FDI attack and recognize FDI
attacker rather than game-theoretic approach?
3. Can we run our experiment in real time for finding out real time security breach?
3.1 FDI Attack Detection
Consider a smart-car network with minimum two cars. At the same time, one of
them will act as sender and another one as the receiver. The sender will collect
sensor information from its environment and store it in array p. Requested by the
receiver, the sender will send ICMP (Internet Control Message Protocol) or UDP
(User Datagram Protocol) packet to the receiver. Let this packet be an array of q.
We can store data of size of i bit into p and q. Important notations adopted in this
paper are listed in Table 1 for easier reading. Due to constant monitoring by the cyber
security analyst, it is possible to compare both p and q. We employ one monitoring
node into this car network. As discussed in Eq. (1), if any bit from p[i] mismatches
to q[i], the monitor can find an FDI attack.
Note that Eq. (1) answers first part of RQ 2 for detecting FDI attack. We may
argue that to understand FDI, we have to realize whether data alteration took place
or not. If we apply game theoretic approach and wait for a cracker to breach the
security, we are definitely going to face property damage in terms hijacking, traffic
jam, accident like issues in vehicle platooning. Let us prove effectiveness of choosing
our approach for real time scenario hypothetically. In game theoretic approach, the
sender will receive requested value first. There is no scope to check the correctness of
150
S. Majumder et al.
Table 1 Summary of notations
Symbol
Meaning
p[i]
q[i]
n−1
p[i] ⊕ q[i] == 1,
i=0
n−1
i=0 p[i] ⊕ q[i] == 0
δG
Sensor data
Sent data
FDI Attack
No FDI Attack
Time required to detect attack in game
theoretic approach
Time required to detect attack in silent listening
approach
Input matrix
Vector for FDI attacker selection
Output vector to choose actual FDI attacker
δP
A[m][n]
b
c[n]
data before it reaches to its destination. So, property damage is compromised. But in
our approach, since the data in-transit will be via our monitoring node, comparison
can take place before it reaches to the receiver. So, in this case, property damage will
not be compromised, which is the main motive for choosing silent listening.
Hypothetically, we define two events H0 and Ha by {H0 : choose δG over δ P } and
{Ha : choose δ P over δG }. Here, δG and δ P denote time required for detecting attack
using game theoretic approach and silent listening, respectively. To proceed further
using silent listening, we should test our hypothesis by following Eq. (1):
H0
δG
<
δ
> P
(1)
Ha
Lemma 1 If vi connects u and ; and vi lies between u and v, time taken to travel
pvi v
puvi
puvi
u vi is less then time required to travel u vi v and vice versa.
Lemma 2 Time required to detect FDI attack using silent spoofing is less than Game
theoretic model.
Proof 1 Let us assume, that initially, δG = δ P = 0. Also assume that all other parameters associated to data in-transit in same network are invariant. As shown in Fig. 1,
we require T (T > 0) seconds to send data from sender to receiver via path uv. Since,
monitor will be in-between sender and receiver, path uv will be decomposed into
puvi
pvi v
puvi
pvi v
u vi v.
∴ weight of uv = T .
Since, in silent listening, vi receives information first, so, weight of uvi < T .
But, in game - based approach, understanding attack requires to cover whole path
u vi v of weight T.
Silent Listening to Detect False Data Injection Attack …
151
Fig. 1 A graph with 3
vertices (sender, monitor and
receiver)
∴ we can conclude that on the basis of path covered and also weight of the sub
paths, δG > δ P .
So, our alternative hypothesis is established and we hence may proceed with silent
listening approach.
3.2 FDI Attacker Recognition
Here we shall get our answer for the second part of RQ 2. From Tables 3 and 4, we
consider input matrix A[m][n] where, m is the number of trial or sent data and n be
the features like p, q, p ⊕ q, FDI, Non-FDI etc. Let b be the vector with values 1, and
0 for separating the trails in two groups FDI attacker and non-attacker, respectively.
To select actual attacker profile, we can use the following equation:
a × b = c , ||A · b|| = |c|
(2)
c[i] > 0 will store the corresponding attacker’s MAC address.
This procedure was accomplished with the help of Algorithm 2.
Proof 2 The proof is given in Sect. 4. From the definition of FDI attack derived in
this paper and the value of a, we can generate normal distribution curve of an FDI
attack as following Fig. 2.
Proof 2 finds the optimal value of a. The above graph shown in Fig. 2 may be illustrated with the help of both Proof 2 and Table 2. Table 2 has mixed data of FDI attack
and non-FDI attack. We may calculate the probability density function (PDF) with
the formula given below:
f (x) = √
1
2π σ 2
e−
1
(x − μ)2
2σ 2
Here, σ , μ represents mean and standard deviation, respectively.
(3)
152
S. Majumder et al.
3.3 Smart-Car Based Test Bed Creation for Sample
Collection
3.3.1
FDI Attack Detection Model
While setting up the test bed, we assume that the network allows all the non-https
contents to be run. We are going to use [25]. We are also going to build the scenario
as stated in [11].
Three machines in a simple Wireless network under same router are connected as
shown in Fig. 3. Let us assume Mc/1 be the sender, Mc/3 be the receiver and Mc/3 is
acting as passive listener (alternatively, sometimes we may address it as ‘monitor’).
Their corresponding IP Addresses are shown in Fig. 3, for example.
Assume that both the sender and receiver are running on Mac operating system
and the monitor is running on Ubuntu Linux. We simply converted the monitor into
a router and issued ARP spoofing on the monitor for sender and receiver. Now, since
the monitor becomes silent listener, it is capable of getting all the data exchanged
between Mc/1 and Mc/2. And being a router, it is able to get the correct sensor data
associated to every machine. Sender and receiver are nothing but the simulated selfdriving cars which are exchanging sensor data. Detailed Simulation of self- driving
car environment is actually not under the scope of the detection algorithm.
Interested readers may follow the process of simulating self-driving car environment as shown in [26]. In M/c-1 and M/c-2, we created a deep learning based
simulated environment with unity as game engine. We trained, validated our cars
with sensor data.
Simulated cars (M/c-1 and M/c-2) are sending and receiving sensor data and the
monitor (M/c-3) is capturing the ICMP packets. ICMP packets are being analyzed in
Fig. 2 Computation of probabilities and percentiles for normal random variables: X ∼ (μ, σ ) [24]
Silent Listening to Detect False Data Injection Attack …
153
Fig. 3 Monitor sniffs data exchanged between two smart-car nodes under lab set-up
real time with actual sensor data. This comparison is very important for understanding
if there exists any difference in the actual sensor data and the obtained ICMP packet
or not. Difference between these two values confirms False Data Injection (FDI)
attack in our experimental set-up.
Table 2 Sample table to compute probabilities and percentiles for normal random variables X ∼
(μ, σ )
FDI?
No. of FDI?
Non-FDI?
No. of Non-FDI
FDI
1
No-FDI
0
FDI
FDI
FDI
FDI
FDI
FDI
FDI
FDI
FDI
FDI
FDI
FDI
Mean
1
1
1
1
1
1
1
1
1
1
1
1
0.5
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
No-FDI
St. dev
0
0
0
0
0
0
0
0
0
0
0
0
0.516397779
154
S. Majumder et al.
Understanding FDI Attack
Snippet for comparing UDP, ICMP packets with actual sensor data:
FOR i = 0, 1, 2, 3, . . . , n − 1 do
IF Sensor _data[i] ⊕ I C M P_ packet[i] == 1
Alert: FDI Attack
ELSE
Default
ENDIF
ENDFOR
Mathematically, the above snippet can also be represented as in Eq. (4):
n−1
i=0
p[i] ⊕ q[i]
== 1, F D I Attack
== 0, N o F D I Attack
(4)
Here, p and q represent actual sensor data and sent ICMP packet, respectively.
As shown in Fig. 4, from the above mathematical derivation of Eq. (4), we can
define FDI attack with the help of state diagram also.
We are collecting data from such five sets in run time. As it is clear that from our
experimental model, we have created post-test only model to study our sample. This
post-test only setting gives us the detailed data set to test our algorithm and give a
valid conclusion over the work.
3.4 Procedure
3.4.1
Comparing Sensor Data and ICMP Packet
Algorithm 1: FDI Detection algorithm (For each Experimental Group)
Fig. 4 State diagram to
understand definition of an
FDI attack
Silent Listening to Detect False Data Injection Attack …
155
STEP 1. INITIALIZATION: Get Actual Sensor data and ICMP packets in two
arrays p, and q, respectively, to solve p[i] ⊕ q[i].
Assuming that initially the system has zero alert, set variable aler t = 0.
STEP 2. FOR i = 0, 1, 2, 3, . . . , n − 1 do
STEP 3. IF p[i] ⊕ q[i] == 1 Generate Alert: FDI Attack
aler t = 1
ELSE
Generate Message: System Scanning Progressing
alert = 0
ENDIF
ENDFOR
STEP 3. STOP
Suppose, there are two finite strings p and q. We have to prove the equality of
these strings. To understand the equality of these two strings, we have to assume that
Pr ((q[i] == p[i])| p[i]) should be 1. We are making this calculation in raw level.
So, we converted our sensor data and ICMP data in binary number system. The state
diagram shown in Fig. 4 helps us to understand how XOR operation is used here to
compare these two values.
3.4.2
Recognizing FDI Attacker
To become an FDI attacker, Pr (M AC[i]|aler t = 1) should be 1. We are using ARP
snooping to understand the ICMP packets exchanged over wireless network. So, we
have advantage to correctly recognize the machines by their MAC addresses.
Algorithm 2: FDI Attacker Recognition Algorithm (For Each Experimental
Group)
STEP 1. INITIALIZATION: Get MAC Addresses of each nodes in an array MAC.
Assuming the system is not under threat, set variable attack = 0
STEP 2. FOR i = 0, 1
IF alert == 1
Declare MAC[i] as FDI attacker
ELSE: Default
156
S. Majumder et al.
ENDIF
ENDFOR
STEP 3. STOP
Remark 1 Note that collaborative security detection for large network is not
scope of the paper. If someone wishes to make a centralized security detection using our algorithm, first they need to setup the experiment accordingly.
Then only she should use the centralized security detection version of it. Theoretically as well as practically we found that paralleling the computation is
appropriate for it.
3.5 Instruments
As per instruction given in [19], we run our simulator in training mode to collect 3
images per frame corresponding to left, right, and center-mounted cameras, steering
angle and speed for each frame. Further based on the model, we generated our
test dataset. From the continuous plot of steering angle, one can easily identify the
existing curve on a path, from the speed and distance covered; one can calculate the
exact length of that curve. So, if M/c-2 wants to get any such sensor information,
M/c-1 may send the information (unit less). Our Algorithms 1 and 2 demand binary
values. So, we are converting the sensor data in binary.
We validated the self-driving model used in both M/c-1 and 2, in total five experimental sets and got the following results as shown in Fig. 5.
We wanted a smooth driving. We trained our model using an ADAM optimizer with a learning rate of 0.0001. We changed this learning rate after each
50 epoch. We applied T-test [27] to assure the statistical significance between
the model Mean Squared Error loss for each experimental group and observed
that ∀x ∈ (1, 2, 3, 4, 5), T _value(x) < critical_value. In set of five experimental groups, we found that there is no significance difference in our self-driving car
model.
We also analyzed the relationship between the self-driving car model given in
[19] and the model simulated in our experimental setup. We analyzed the T-value of
the sample means of the steering angle obtained in both our model and the former.
We found that the two models are statistically significant, since ∀x ∈ (1, 2, 3, 4, 5),
T _value(x) < critical_value, given model in [19].
Silent Listening to Detect False Data Injection Attack …
157
3.6 Data Analysis Technique
The basic difference between an FDI attacker and a non-FDI attacker were made on
the basis of the mean and standard deviations as shown in Table 3.
Definition 1 We are defining an attack (in our case, FDI) with mean value as one
and zero standard deviation, and a non-attack is defined as both zero mean and zero
standard deviation.
For our dataset, typically we can compute, by the law of Binomial Distribution
[28], for 100 numbers of trials, Pr (Success f ul_Attack) with mean value, μ =
np = 50.
Fig. 5 Validation of our
simulated smart car models
Table 3 Sample data collected for FDI attack and non-FDIs
Result
FDI
Mean
St. dev
1
1
1
1
1
1
1
1
1
0
Non-FDI
0
0
0
0
0
0
0
0
0
0
158
S. Majumder et al.
Fig. 6 Performance of
Algorithms 1 and 2
4 Results and Discussion
From Fig. 6, we find that both Algorithms 1 and 2 are capable of successfully detecting
100% FDI attacks, as well as recognizing the attacker in run time.
We apply T-Test to check the statistical significance of the test at initial level. We
also apply Chebyshev’s inequality.
As shown in Table 4, we can detect the attack with 100% success rate.
Successful detection of attack or no-attack = 100%.
Successfully recognizing the data value sent over network = 100%.
Successfully recognizing the attacker node = 100%.
From the literature study, it is quite obvious to understand that game-based attack
detection does not guarantee prevention from accidents or damage in vehicle platooning. But, it is clear from Lemmas 1 and 2, that since the attack detection takes
place before successful attack on the target machine, the present approach may save
property damage in FDI attacks compared to the game-based one.
Table 4 Data for understanding FDI
Trial
p = Sensor
q = Sent
Data
Data
i=1
i=2
:
100
101
:
101
101
:
p⊕ q
Comparison
Meaning
110
111
:
q= p⊕q
q= p⊕q
:
FDI
No FDI
:
Silent Listening to Detect False Data Injection Attack …
159
4.1 Examining Correctness of Algorithms 1 and 2
Let us check here the correctness of our algorithm (FDI attack detection algorithm).
Assume, Z takes two values 0 and 1 as follows Eq. (5):
1, i f p[i] ⊕ q[i] = 1
Z=
0, i f p[i] ⊕ q[i] = 0
(5)
Assume, Y denotes the attack occurs. Since, we know in our case, the expected
value of Y is Eq. (6),
E[E[Y |Z ]] = E[Y ]
(6)
We can claim that the working principle of our algorithm for comparing actual sensor
value and the sent values makes the algorithm correct.
Proof 3
L .H.S = E[E[Y |Z ]]
=
Pr (Z = z)
y · Pr (Y = y|Z = z)
z
=
=
=
y
z
y
z
y
y · Pr (Y = y|Z = z) · Pr (Z = z)
y · Pr (Y = y ∩ Z = z)
y · Pr (Y = y)
y
= E[Y ]
= R H S(H ence Pr oved).
Since, as we are using binomial distribution, we will now use the Chebyshev’s
inequality.
For any random variable, its variance to be given as Eq. (7):
V ar [X ] = E[X 2 ] − (E[X ])2
(7)
Assume X denotes a machine to be an attacker i.e. it is associated to the attacker
recognition algorithm (Algorithm 2). Therefore, if we want to know what is the
expectation that an FDI attack launched and the selected machine is an attacker, can
be given by Eq. (8):
V ar [X + Y ] = V ar [X ] + V ar [Y ] + 2Cov(X, Y )
(8)
160
S. Majumder et al.
Clearly, we understand that, this relationship between X and Y is actually helping us
to identify the importance of M/c-3 i.e. the monitor.
Let us understand first the dependency of X and Y. If we further expand the
covariance, we will get Eq. (9):
Cov(X, Y ) = E[X · Y ] − E[X ] · E[Y ]
(9)
If X and Y are completely independent then Cov(X, Y ) should be 0. Or, we can
say, as in Eq. (10) that:
E[X · Y ] = E[X ] · E[Y ]
(10)
Assume the situation where we have to recognize an attacker particularly when an
attack takes place. In that case, if X and Y are not dependent, we might get Eq. (11),
E[X · Y ] − E[X ] · E[Y ] = 0
(11)
We may argue if, X and Y are dependent, Cov(X, Y ) becomes 0 · 75. We assume
that the expectation is ranging between 0 and 1. So, it is clear that to establish the
need of knowing the MAC address of an attacker will only be applicable iff X and Y
are dependent. In our case, the role of monitor is backed with the working principle
of ARP Spoofing where it constantly monitors ICMP packets and sensor value with
respect to particular MAC address. Thus we can defend the usage of ARP spoofing
in Algorithm 2 for attacker recognition.
Proof 4
E[X · Y ] =
x
=
x y · Pr (X = x) · Pr (Y = y)
y
x · Pr (X = x) ·
x
y · Pr (Y = y)
y
= E[X ] · E[Y ]
When X and Y are independent.
Proof 5 In our case,
= 1, i f Pr (X ) = 1
X
= 0, i f Pr (X ) = 1 − p
(12)
Assume,
E[X ] = p
(13)
E[X 2 ] = 1 − p
(14)
and
Silent Listening to Detect False Data Injection Attack …
161
Therefore, for successful attacker recognition with probability 1 and unsuccessful
attacker recognition with probability 0 lead the value of Var[X] to be Eq. (15):
= 1, f or Pr (X ) = 1
V ar [X ]
= 1, i f Pr (X ) = 0
(15)
We derive Eq. (15) by applying the formula Eq. (16),
V ar [X ] = p − p 2
(16)
We shall consider the situation with the help of another example. Out of two given
ranges of MAC addresses, identifying a particular machine is denoted with a >0.
Therefore, applying two-tailed test using Chebyshev’s inequality, we get Eq. (17):
Pr (|X − E[X ]| ≥ 0) ≤
V ar [X ]
a2
(17)
From the rule of Binomial distribution, X ∼ B(n, p).
For each successful attack, in every single trial, n will be 1, for knowing the MAC
address correctly of a particular machine, p will be 0.5.
Therefore,
E[X ] = np = 0 · 5
(18)
Therefore,
V ar [X ] = E[X 2 ] − (E[X ]2 ) = 1 − p 2 = 1 − 0 · 25 = 0 · 75
(19)
Finally, we can solve Eq. (17) by Eq. (18)
V ar [X ]
0 · 75
= 0 · 75
=
2
a
1
(20)
Proof 6 From Table 5, we can write:
11
A=
00
Table 5 Data for recognizing attacker
Trial
Comparison
i=1
i=2
:
q = p⊕q
q = p⊕q
:
(21)
Meaning
Sender’s MAC address
FDI
No FDI
:
8:0:27:ba:fd:31
8:0:20:af:bd:35
:
162
S. Majumder et al.
From Eq. (21), we get:
1 1 1 · 1 + 1 · 0 + 1 · 1 + 1 · 0 2
=
· 10 =
0 · 1 + 0 · 0 + 0 · 1 + 0 · 0 0
00
(22)
Clearly, from the output matrix c stated in Eq. (22), we find, for cases c[i] =
0, there was no attack. From Table 5, we took two features denoting comparison
and meaning. Based on c[i] > 0 or more precisely, c[i] = 2, we can choose the
corresponding MAC address of the attacker, i.e. the sender for a particular trial.
To minimize approximation error, we apply Chernoff Bounds.
Theorem 1 Let X 1 , X 2 , . . . , X n be independent random variables with Pr (X i =
= 0) = 21 .
1) = Pr (X i n
Xi .
Let, X = i=1
−a 2
Then Pr (X ≥ a) ≤ e 2n .
In our case, we are not interested in mean = 0. Neither we are interested in the
events those lead to c[i] ≤ 0. So, we shall use one tail test for a > 0 values only.
Since, we
√ are interested in the non-zero terms of A[m][n], we need to calculate
Pr (|c1 | > 4mlnn).
Pr (|c1 | >
√
4mlnn) ≤ e−
4mlnn
replacing a 2 with 4 mlnn
2k
(23)
where, k is the term corresponds to 1 in A[m][].
√
From the example of Proof 6, we find, Pr (|c1 | > 4mlnn) ≤ e− 4mlnn
2∗k
−2ln2
≤ e− 4∗1ln2
≤
e
≤
0.25.
If
we
tail
bound
both
the
sides,
then
we
shall
get
2∗1
Pr (|c1 | > a) = 2 ∗ 0.25 = 0.50.
4.2 Brief Answers for RQs
From the experiment, data collection and analysis of the results we derive the following answers of our RQs:
1. In Fig. 5, we already proved the equivalence of our test-bed with existing deep
learning based self-driving cars. Hence, we conclude that yes, we can simulate
real time, deep-learning based smart car network as our test-bed.
2. In Sects. 3 and 4, we showed the importance of ARP spoofing, the analysis of
Algorithms 1 and 2; also we analyzed the dependencies of detecting attack and
recognizing attacker in present work. Hence, we may argue that ARP spoofing
based algorithm could be an alternative of game based algorithms. In addition,
Lemmas 1 and 2 prove the importance of our algorithm over game-based solutions.
Silent Listening to Detect False Data Injection Attack …
163
3. We carry out our experiment under real time set-up. We launch and detect FDI
attacks in real time. Thus, we can argue that RQ3 is true.
5 Conclusion
The chapter shows how ARP spoof based algorithms can detect False Data Injection
attacks and recognize the attacker successfully in a simulated smart-car test bed. It
gives idea to beginners of test bed creation with deep learning algorithm for selfdriving cars of automation level 5. It discusses how to launch an FDI attack in order
to find security breach in the system. It also shows how to recognize the FDI attacker
using ARP spoofing which is a hacker’s tool. Finally, it argues the applicability of
ARP spoofing, over game theory based solution.
We analyzed 100% FDI attack detection and attacker recognition using Algorithms 1 and 2, presented in this paper.
In spite of all of them, we have some limitations, and we want to work on them
as future work.
We selected less number of computers in real life setup. Initially, we prepared five
different experimental groups with two MAC machines and one LINUX machine.
Then we gathered data and tested Algorithms 1 and 2, under both wireless and hybrid
combination of wired and wireless) network.
Since, our motive is also to deal with FDI attacks in smart car platooning, then
we prepare next set of experiment and find some interesting observations like:
• applying ARP spoofing from monitor allows to track other suspicious node to
apply ARP spoof based attacks.
• IP forwarding can immediately stop serving any suspicious or malicious node from
the network.
Even after carrying out the experiment with high configuration machines, we
observed that the analyzed and produced data are huge. As a solution we decide to
paralleling the task and use high performance computers to monitor more than two
nodes, in future.
Since detailed study regarding smart cars is out of the scope of the chapter, we
focus more on the behavioral components of the modern smart cars. We shall work
more on our present test bed as part of future work.
ARP spoof is a strong tool used by hackers. On the other hand, cyber security is
such a subject that to know the attacks we have to know the hackers’ techniques. In
real time situation, different vulnerability analysis or threat analysis tools are there
for system testing, but instead applying them, we directly used it to compare sensor
data. In reality, no theoretical books tell us about direct approach to launch such
attacks.
Aim of this chapter is basically to encourage the students of cyber security to
identify security breaches, come up with solution with the help of hands-on knowledge of hacking like ARP-Spoofing based tool, under the supervision of experts and
164
S. Majumder et al.
security analysts. Finally, we conclude that from security analyst’s point of view, and
sensor network point of view, we may apply such readily available tools to defend
security breaches.
Acknowledgements We are thankful to Information Security lab, NIT Agartala. The fellowship of
the first author (research scholar) working under the research work is funded by the MHRD, Govt.
of India.
References
1. A. Humayed, J. Lin, F. Li, B. Luo, Cyber-physical systems security—a survey. IEEE Int. Things
J. 4(6), 1802–1831 (2017)
2. Q. Xu, B. Wang, F. Zhang, D.S. Regani, F. Wang, K.J.R. Liu, Wireless AI in smart car: how
smart a car can be? IEEE Access 8, 55091–55112 (2020). https://doi.org/10.1109/ACCESS.
2020.2978531
3. K. Koscher et al. Experimental security analysis of a modern automobile, in Proceedings of
the IEEE Symposium Security Privacy (SP) (Oakland, CA, USA, 2010), pp. 447–462
4. T. Hoppe, S. Kiltz, J. Dittmann, Security threats to automotive CAN networks’ practical examples and selected short-term countermeasures. Reliab. Eng. Syst. Saf. 96(1), 11–25 (2011)
5. B. Biswas, Bidisha, Analysis of false data injection in vehicle platooning. All Graduate Theses
and Dissertations 4038 (2014). https://digitalcommons.usu.edu/etd/4038
6. B. Littlewood, S. Brocklehurst, N. Fenton, P. Mellor, S. Page, D. Wright, J. Dobson, J. McDermid, D. Gollmann, Towards operational measures of computer security. J. Comput. Secur. 2(3),
211–229 (1993)
7. Z. Liu, J. Tang, Z. Zhao, S. Zhang, Adaptive neural network control for nonlinear cyber-physical
systems subject to false data injection attacks with prescribed performance. Philos. Trans. R.
Soc. A 379(2207), 20200372 (2021)
8. N.A. Carreon, L. Sixing, R. Lysecky, Probabilistic estimation of threat intrusion in embedded
systems for runtime detection. ACM Trans. Embedded Comput. Syst. (TECS) 20(2), 1–27
(2021)
9. H. Wu, W. Wang, A game theory based collaborative security detection method for Internet of
things systems. IEEE Trans. Inf. For. Secur. 13(6), 1432–1445 (2018)
10. https://en.wikipedia.org/wiki/Self-driving_car
11. https://www.veracode.com/security/arp-spoofing
12. M. Wolf, A. Weimerskirch, C. Paar, Security in Automotive Bus Systems (2004)
13. X. Yang et al., A novel en-route filtering scheme against false data injection attacks in cyberphysical networked systems. IEEE Trans. Comput. 64(1), 4–18 (2015)
14. L. Xiao, Y. Chen, W.S. Lin, K.J.R. Liu, Indirect reciprocity security game for large-scale
wireless networks. IEEE Trans. Inf. For. Secur. 7(4), 1368–1380 (2012)
15. Y. Han, T. Alpcan, J. Chan, C. Leckie, B.I.P. Rbinstein, A game-theoretic approach to defend
against co-resident attacks in cloud computing: preventing co-residence using semi-supervised
learning. IEEE Trans. Inf. For. Secur. 11(3), 556–570 (2016)
16. L. Xiao, J. Liu, Q. Li, N.B. Mandayam, H.V. Poor, User-centric view of jamming games in
cognitive radio networks. IEEE Trans. Inf. For. Secur. 10(12), 2578–2590 (2015)
17. F. Li, Y. Yang, J. Wu, Attack and flee: game-theory-based analysis on interactions among nodes
in MANETs. IEEE Trans. Syst. Man Cybernet. Part B Cybernet. 40(3), 612–622 (2010)
18. A. Buyval, A. Gabdullin, R. Mustafin, I. Shimchik, Realtime vehicle and pedestrian tracking
for didi udacity self-driving car challenge, in 2018 IEEE International Conference on Robotics
and Automation (ICRA) (2018)
19. https://github.com/camigord/Self-Driving-Car-Nanodegree/tree/master/P3-BehavioralCloning
Silent Listening to Detect False Data Injection Attack …
165
20. http://www.image-net.org/
21. A. Krizhevsky, I. Sutskever, G.E. Hinton. Imagenet classification with deep convolutional
neural networks, in Advances in Neural Information Processing Systems (2012), pp. 1097–
1105
22. M. Montemerlo, J. Becker, S. Bhat, H. Dahlkamp, D. Dolgov, S. Ettinger, D. Haehnel, T.
Hilden, G. Hoffmann, B. Huhnke et al., Junior: the stanford entry in the urban challenge. J.
Field Robot. 25(9), 569–597 (2008)
23. J. Wei, J.M. Snider, J. Kim, J.M. Dolan, R. Rajkumar, B. Litkouhi, Towards a viable autonomous
driving research platform, in Intelligent Vehicles Symposium (IV), 2013 IEEE (IEEE, 2013),
pp. 763–770
24. https://homepage.stat.uiowa.edu/~mbognar/applets/normal.html
25. RFC 826, An Ethernet Address Resolution Protocol (Network Working Group, Plummer, 1982)
26. https://github.com/udacity/self-driving-car-sim
27. Student, The probable error of a mean. Biometrika 6(1), 1–25 (1908). JSTOR. www.jstor.org/
stable/2331554
28. https://en.wikipedia.org/wiki/Binomial_distribution
Taxonomy of UAVs GPS Spoofing
and Jamming Attack Detection Methods
A. Sabitha Banu
and G. Padmavathi
Abstract The fast evolution of the Unmanned Aerial Vehicle (UAV) has added
a great deal of ease to our lives. Specific characteristics of UAV networks, such
as node mobility and network design, differ. Security is a significant issue with
UAVs since they are used for such delicate tasks. Both the business and academia
are interested in the security of unmanned aerial vehicle networks, and as a result,
to keep data safe from hackers and fictional actions by unauthorized users, some
effective methods must be employed. Numerous threats may attack such networks
with the intent of jamming communication, interfering with network functioning, or
injecting incorrect data. Since unmanned aerial vehicles (UAVs) depend on the Global
Location System (GPS) for positioning and navigation, they are susceptible to GPS
spoofing and jamming attacks. Numerous studies are being conducted to improve the
robustness of UAV routing protocols and detection mechanisms through the use of
Machine Learning (ML), Deep Learning (DL), and Computational Intelligence (CI)
techniques. These studies also aim to improve battery life, network performance, and
security against attackers. This chapter discusses the different classification of UAVs
and their applications, a Statistics report of the current UAV market, some design
considerations to build a UAV, taxonomy of UAV routing protocols, and a study of
ML, DL, and CI methods used to detect the most predominant attack called GPS
Spoofing and Jamming attacks.
Keywords UAV · GPS spoofing attack · Jamming attack · Taxonomy · ML/DL
detection methods
1 Introduction
An unmanned aerial vehicle (UAV) is a plane that can fly without a human pilot.
Instead, it is remotely piloted or autonomously flown by on-board computer systems.
UAVs are also known as drones [1]. A wide variety of applications and industries
A. S. Banu (B) · G. Padmavathi
Avinashilingam Institute for Home Science and Higher Education for Women, Coimbatore, India
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_10
167
168
A. S. Banu and G. Padmavathi
Fig. 1 UAV applications
are possible with UAVs. Drone applications [2] will spread into many areas of life,
which are illustrated in the below system map Fig. 1.
Furthermore, they also assisted in covid-19 pandemic situations without any
human contact. Sometimes drones are also used in locations that were not able to
reachable. Earlier, UAVs were mainly employed for military purposes. They carry
out operations that put human pilots in danger. Recently, however, UAVs have been
found increasing for civilian uses. Some civilian operations are search and rescue,
policing, and inspecting military movements, strategic activities, or environmental
monitoring [3].
1.1 Motivation Based on the Statical Reports
According to a CNBC report [4], the UAV industry grew fast, from $100 billion in
2020 to $1.5 trillion by 2040, shown in Fig. 2.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
169
Fig. 2 UAV market forecast
1.2 Classification of UAVs
There are different types of UAVs are available in the market. They have differed
in terms of Size, Payload, Efficiency, Complexity, and Usage. UAVs vary in size,
weight, payload, speed, range, endurance, and electrical and mechanical design.
Some of them are [5]
• Single Rotor drones
• Multirotor drones
–
–
–
–
•
•
•
•
•
•
•
•
•
•
•
•
Tricopter (3 rotors)
Quadcopter (4 rotors)
Hexacopter (6 rotors)
Octocopter (8 rotors)
Fixed-wing drones
Fixed-wing hybrid drones
Small drones
Microdrones
Tactical drones
Reconnaissance drones
Large combat drones
Non-combat large drones
Target and decoy drones
GPS drones
Photography drones
Racing drones.
There are several numbers of UAVs used for both military and commercial
purposes based on their usage. Multiple UAVs, sensors, base stations, and data transmission connections comprise UAV networks. The majority of UAV missions need
three-dimensional interaction with the operator, necessitating many on-board control
sensors. The transmission and management of real-time data by a UAV need reliable
and coordinated communication connections between sensors. The primary control
170
A. S. Banu and G. Padmavathi
components are physical infrastructure (external hardware), computer systems
(internal hardware), and non-physical software. The payload, operator, data connections, and support components are all similar to UAVs. Several data connections may
be established during flight, even in autonomous mode, with one or more Ground
Control Stations (GCS), ground-based antennas, Mobile Ground Units, or other UAV
(e.g., in the case of swarming). A communication connection may be either continuous and dedicated (like Wi-Fi or Bluetooth) or discrete (like a cable or satellite link).
UAV networks rely on data flow via sensors, connections, avionics, and hardware
infrastructures like any other IT network or operating system [5].
1.3 Design Considerations of UAV
Due to their unique features, UAV networks need many critical design considerations.
Several factors to consider before building UAV networks include the following [6]:
•
•
•
•
•
•
•
•
Topology
mobility
latency
frequent link disconnection
prediction, flight formation
collision avoidance
combat with external disturbances and
scalability.
UAV networks need a high degree of scalability, flexibility, and robust routing
protocols because of the fast changes in topology and activities during network
operation. Routing protocols of UAV networks are examined in the coming section.
In most cases, UAV networks use three distinct modes of data transmission. They
are unicasting, broadcasting, and multicasting. In most routing systems, packets are
delivered via which to get to the desired location via series of one-to-one connections
from the source node. Broadcast data transmission entails transmitting information
in the form of packets from a specific source node to all destinations nodes in the
network region. Probabilistic routing, network coding, and other methods are built
on multicast, which is a node that distributes data packets to a group of target nodes
in advance.
Earlier, UAV networks were tested using the routing technologies employed in
MANET and VANET. However, it failed due to the inability of the system to adapt the
unmanned aerial vehicles’ high level of mobility, rapidly changing network architecture, and poor communication links, routing overhead, bandwidth, scalability.
It may be addressed by either improving existing routing protocols, hybridizing
different routing protocols, or replacing them with new ones [7]. Figure 3 illustrates
the taxonomy of UAV routing protocols, from conventional to AI-enabled routing
protocols.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
171
Fig. 3 Taxonomy of UAV routing protocols
2 Taxonomy of UAV Routing Protocols
Routing protocols for unmanned aerial vehicles (UAVs) are divided into two distinct
groups: (i) network architecture-based and (ii) data-forwarding-based, and again,
in turn, network architecture-based is subdivided into “topology-based routing,
position-based routing, hierarchical routing, SI based routing, probabilistic routing,
and AI-enabled routing”. Data forwarding-based is subdivided into “deterministic
172
A. S. Banu and G. Padmavathi
routing, stochastic routing, and social networking-based routing” [6, 8, 9]. Subdivisions of the routing as mentioned earlier and their protocols are explained in detail
below.
2.1 Topology Based Routing
Network packets are routed using topology-based routing techniques. Topologybased routing systems use IP addresses to define nodes. The subdivisions of topologybased routing protocols are:
•
•
•
•
Static Routing Protocol
Proactive Routing protocol
Reactive Routing protocol
Hybrid routing protocol.
2.1.1
Static Routing Protocol
Static routing techniques have fixed routing. The UAVs can’t update or change their
routing database while in flight. It is also a fault-tolerant routing protocol.
Data Centric Routing (DCR)
Data-centric routing uses data to work. For one-to-many communications, these
routing methods may be utilized. This protocol performs better in clustered
environments.
Loar Carry and Deliver Routing (LCAD)
Data transfer from the source ground station to the destination is safe using this
method while maintaining high throughput. However, because of vast distances, the
method’s primary flaw is a significant transmission latency. A multi-UAV system
may very well be utilized to minimize data transmission latency.
Multi-level Hierarchical Routing (MLHR)
Several clusters in a hierarchical network may carry out distinct functions. MLHR
may have a flat basis. The hierarchical design expands the network’s operating area.
UAV networks are clustered, with just the cluster head (CH) connecting to other
cluster heads and the ground node.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
2.1.2
173
Proactive Routing Protocol
Proactive Routing protocol utilizes routing information to store all network routing.
The nodes regularly update and exchange their routing tables. As a result of changes
in topology, it must be updated in the tables. The main benefit of PRP is that it is
constantly up to date. All communication nodes must exchange routing messages to
keep the routing tables updated. However, it uses an excessive amount of bandwidth
which ruins the network. Changes in the topology of a network affect how quickly
it reacts, resulting in latency.
Optimized Link State Routing (OLSR)
OSLR is a mobile ad hoc routing protocol. Due to its proactive character, a linkstate algorithm’s stability is inherited by the protocol. OLSR is a link-state protocol
optimization for mobile ad hoc networks. OLSR is intended to operate fully decentralized, with no central authority. Because each node transmits loss of control signals
regularly, the protocol allows for tolerable levels of damage. When nodes exchange
control messages, overhead is generated. Numerous additional routing protocols have
been suggested based on the OLSR technique, including D-OLSR, M-OSLR, and
CE-OSLR.
Directional Optimized Link State Routing (D-OLSR)
This method reduces the number of multipoint relays by utilizing directional
antennas. If the distance from the origin to the destination is more than Dmax /2,
the node uses the DOLSR technique. The OLSR with an omnidirectional antenna is
utilized for distances less than Dmax /2. In this way, we reduce the latency and network
finding overhead. However, the overhead in UAV networks is still too high.
Multidimensional Perception and Energy Awareness OLSR (MPEAOLSR)
OLSR has been enhanced to take node connection time, link-layer congestion level,
and node residual energy into account when determining the best route to take.
Data transmission success rates are improved, packet loss is reduced, and end-to-end
latency is minimized as a result.
Dynamic Dual Reinforcement Learning Routing (DDRLR)
For multi-service wireless mesh networks, this research explores an independent
Quality of Experience (QoE) methodology. Asymmetric forward/backward reinforcement learning techniques are used for each mesh node to dynamically change
174
A. S. Banu and G. Padmavathi
their routing strategies to optimize the perceived QoE for each network flow by the
user. A new source rate adaptation technique is used in conjunction with the routing
methods to match the transmitting rate to the available network capacity.
Destination Sequence Distance Vector Routing (DSDV)
The Bellman-Ford method is modified somewhat in DSDV routing. DSDV is a
proactive technique for routing data transmission. DSDV uses incremental and fulldump update packets. Network topology changes are signaled by sending incremental
packets, reducing network overhead, and not eliminating it. To prevent network loops,
DSDV routes are numbered sequentially. However, updating the route takes much
bandwidth.
BABEL
A distance vector technique that avoids loops, BABEL. It is better suitable for
version 4 and version 6 of Internet protocol networks. The BABEL can improve
loop-free concurrence rapidly. BABEL utilizes a metric to determine the shortest
route. Because BABEL updates the periodic routing database, it produces additional
traffic when the network topology changes. BABEL fails in UAV networks based on
datagram loss rate and average outing time.
Better Approach to Mobile Ad Hoc Network (BATMAN)
Ad hoc networks’ BATMAN is a relatively recent proactive routing system.
BATMAN actively maintains the presence of all communication nodes via single-hop
and multi-hop. Next-hop neighbors may be utilized to establish communication with
the target node. The BATMAN algorithm is advantageous for finding the optimum
subsequent hop for each position. Because BATMAN does not compute the whole
path, it is rapid. BATMAN has excellent data rate and packet loss characteristics.
The BATMAN packets are small because they can only hold so much data. Because
the packets lack route information, BATMAN cannot spread.
Optimized Link Routing with Expected Transmission Count (OLSR-ETX)
The OLSR-ETX outperforms the conventional OLSR packet transfer, delay, and
overhead.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
175
Link-Quality and Traffic-Load Aware Optimized Link-State Routing
(LTA-OLSR)
LT-OSLR uses statistics on the received signal strength to find the link quality.
Multidimensional Perception and Energy Awareness OLSR (MPEAOLSR)
Node connection time, residual energy, and link-layer congestion are all factors
in improved OLSR route selection. Reduces packet drop and improves data
transmission success rates while reducing overall delay.
Cluster Head Gateway Switch Routing (CGSR)
This is a table-based algorithm for clustered networks. In this approach, each node
keeps two tables: one for cluster membership and one for routing. It is quick to find
the route. Choosing the cluster head does add to the routing protocol’s complexity.
It also does not apply to flat mesh networks.
Wireless Routing Protocol (WRP)
It includes the routing table network view. The nodes maintain several tables to
improve performance: Distance Table (DT), Route Table (RT), Link Cost Table
(LCT), and Message Retransmission List (MRL). It finds the shortest route and
converges faster than DSDV. A new connection break technique is enabled by
monitoring update messages. Cost is significant table update overhead.
Topology Broadcast Based on Reverse Path Forwarding (TBRPF)
It uses reverse path forwarding. It utilizes minimum-hop trees instead of shortest-path
trees to save on overhead. To minimize routing message cost, it compares the past
and current network states instead of the entire network status. However, it seldom
finds the shortest route.
2.1.3
Reactive Routing Protocol
They generate demand-based routing data. It implies that finding a path can be
performed when a transmission session is required. This method reduces overhead,
particularly in low-traffic areas. If a route fails, constructing another route takes a
while. Listed below are part of prominent reactive routing protocols.
176
A. S. Banu and G. Padmavathi
Dynamic Source Routing (DSR)
This routing system saves a complete route from one node to the next. It has a
lower overhead than AODV, OSLR, and (TORA). The major disadvantage is the
route-finding time.
Ad Hoc on Demand Distance Vector (AODV)
It is a standard routing method for storing the next node in a routing table’s next hop,
maximizing network bandwidth efficiency and reducing routing overhead. However,
route determination takes too lengthy, causing network congestion.
Radio Metric Ad Hoc On-Demand Distance Vector (RM-AODV)
IEEE 802.11s suggested this protocol. RM-AODV eliminates the route determination
complexity from the top layers, allowing them to view all UAVs in one hop. The
protocol’s route cost measure indicates connection quality and resource consumption
when a frame is transferred across a link.
Dynamic Topology-Multipath AODV (DT-MAODV)
To reduce the frequency of route rebuilding, this routing protocol utilizes a route that
is both linked and disjointed to create a large number of alternative routes. When
compared to AODV, it reduces packet loss and reduces end-to-end latency.
Associativity-Based Routing (ABR)
The present routing method utilizes a connectivity database to maintain track of
network connections. Other reactive routing methods need more route reconstructions. This risk rises with increased mobility and traffic.
Signal Stability-Based Adaptive Routing (SSA)
When a steady connection is not available, this method chooses unstable connections.
It prioritizes signal strength and connection reliability. This method’s route setup time
is lengthy.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
177
Message Priority and Fast Routing (MPFR)
In order to expedite data transmission, this protocol utilizes priority bits in the
decision-making process. Using this method, the essential packets are processed
first, which reduces the processing time for lower-priority ones.
Dynamic Backup Routes Routing Protocol (DBR2P)
When a link fails, this routing method includes a backup node mechanism to quickly
reconnect, allowing backup routes to be discovered when the original one is lost.
Dynamic MANET On-Demand (DYMO)
The Route Request (RREQ) packets are broadcast along with the location data using
this method. Compared to AODV, it has a reasonably low overhead, but it takes
longer to set up routes.
Time Slotted On-Demand Routing (TSOR)
Solves network congestion using this routing method. Each UAV has its time slot,
reducing the communication overhead between UAV pairs.
Reactive-Greedy-Reactive (RGR)
For UAVs, a version of AODV with fewer hops was suggested. However, the shortest
route is not always the best.
Modified-RGR (M-RGR)
Routes in RGR that are more reliable and stable. Packet forwarding UAVs seek high
link dependability. This method uses GPS data to determine the distance between
two points. Data transmission between two nodes traveling in opposing directions is
lost. Otherwise, the dedicated route uses the AODV protocol.
AODV Security Extension (AODVSEC)
A lack of security in the AODV routing protocol does not render it unreliable on an
open network. For security, AODVSEC proposes to extend the scope of AODV with
less computation time.
178
A. S. Banu and G. Padmavathi
Trusted AODV (TS-AODV)
A protocol is described that allows nodes to participate in routing depending on their
trust rating. If a node’s trust value exceeds a threshold, it is recognized as trustworthy
and permitted to participate in routing.
2.1.4
Hybrid Routing Protocols
Hybrid routing methods are developed to reduce overhead issues in proactive and
reactive protocols. In contrast, Proactive Routing Protocol has a high overhead of
control messages and requires more time to find routes. Intra-zone routing utilizes
Proactive Routing Protocol, whereas inter-zone routing uses Reactive Routing
Protocol. Some of the hybrid routing protocols for UAV networks are listed below.
Zone Routing Protocol (ZRP)
Make use of the concept of clustering. For inter clustering, it uses proactive, and intraclustering uses a reactive approach. This technique reduces the processing time and
overhead of the route finding. Maintaining information is nevertheless tricky for
dynamic nodes and connection behavior.
Temporarily Ordered Routing Algorithm (TORA)
Networks with many hops utilize this routing technique where routers only save
information about nearby nodes. However, it restricts control message transmission
in highly dynamic networks, reducing UAV network efficiency.
Hybrid Wireless Mesh Routing Protocol (HWMP)
It aids in route selection. In multi-hop networks, HWMP is utilized for video
surveillance.
On-Demand Routing with Boids of Reynolds Protocol (BR-AODV)
Routing based on demand and the Boids of Reynolds technique is used in BR-AODV
to guarantee routing and connectivity during data transfer. Boids of Reynolds are used
in this protocol to preserve connection when the UAV’s dynamic topology changes.
The AODV protocol is used to transmit messages between unmanned aerial vehicles
(UAVs) when needed because it enables the UAV nodes to acquire routes if required.
The number of route-finding launches may be reduced by using BR-AODV instead of
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
179
a conventional satellite. Packet delivery ratio, delay, and packet drop are all improved
over AODV.
Link Estimation-Based Preemptive Routing (LEPR)
The AODV routing protocol is the foundation for this protocol. In developing LRPR,
the link stability measure is used, with GPS providing node position information.
Additionally, the quality of the connection, the degree of safety, and the likelihood of
node mobility are considered. Multiple link-disjoint routes are explored throughout
the route discovery process. LEPR minimizes broken pathways and end-to-end
latency.
Reactive Flooding Routing (RFR)
RFR is suggested for scenarios involving the use of technology in farming in which
abrupt climatic changes significantly impact the quality of the crops and the farming
methods. UAVs are equipped with a specialized sensor that transmits data to farmers.
The reactive method outperforms the proactive approach in terms of packet delivery
ratio.
Sharp Hybrid Adaptive Routing Protocol (SHARP)
SHARP is a hybrid routing system that adapts to changing network and traffic
characteristics. SHARP allows applications with varying needs to adjust routing
layer performance. SHARP allows applications to explore this area in networks with
dynamic traffic patterns, node degrees, and mobility rates. SHARP outperforms in
terms of packet overhead, loss rate, and jitter.
2.2 Position-Based Routing
With GPS, the present algorithms determine the best route depending on the user’s
location. For instance, the following node may be chosen as a result of proximity
to the destination node. These techniques’ primary drawback is their reliance on
precise positioning and tracking devices. This routing is ideal for networks of highly
dynamic UAVs. The following section contains a summary of the protocols.
180
2.2.1
A. S. Banu and G. Padmavathi
UAV-Assisted VANET Routing Protocol (UVART)
Algorithms that take connection and traffic density into account while routing traffic.
UVAR is controlled by the following four factors: traffic density, node distance,
connection, and vehicle dispersion. The Dijkstra algorithm is used to locate the
source and destination. When using UVAR to predict the number of automobiles in
a certain area. When the network is on the ground, UVAR may serve as a relay. It
also offers road signs and assists traffic management.
2.2.2
Connected-Based Traffic Density Aware Routing Protocol
(CRUV)
The cars exchange periodic HELLO packets. These exchanges are used to identify
the most linked segments. UAVs share this data with other nodes to optimize route
choices. If a segment is linked, the source vehicle chooses the UAV to send data to.
Its main benefit is that UAVs discover it when the present vehicle cannot locate the
linked section.
2.2.3
UAV-Aided Vehicular Networks (UAV-VN)
A network’s route availability is dependent on vehicle density and collaboration.
It improves route connectivity. SCF allows UAVs to help ground vehicles in data
transmission to roadside units (RSU).
2.2.4
UAV Relayed Tactical Mobile Ad Hoc Networks
(UAVRT-MANET)
The UAV-aided relay node creates a temporary route in MANET. The relay node
temporarily joins partitioned networks and offers backup routes.
2.2.5
Predictive-Optimized Link State Routing Protocol (P-OLSR)
P-OLSR used in UAV networks to predict connection quality using GPS. Geographical locations are shared between the nodes by exchanging HELLO packets. Using
this method, each node knows its neighbors’ positions. The protocol can be utilized
in high-speed UAV networks.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
2.2.6
181
Predictive Routing for Dynamic UAV Networks (PR-DUAV)
The PR-DUAV routing system uses predicted intermediate node positions to choose
paths. The route selection criteria of Dijkstra’s shortest path method are updated
to include anticipated pairwise distances. It lowers calculation time and improves
the delivery route. The PR-DUAV routing method extends network connection
by eliminating links that are likely to be severed owing to communication range
exceeding.
2.2.7
Location-Aware Routing for Opportunistic Delay-Tolerant
Networks (LAROD)
LAROD is a delay-tolerant geographical routing system based on greedy forwarding
and store carry and forward. It has a good network delivery ratio. LAROD is
appropriate for mini-drones but consumes too much energy.
2.2.8
Deadline Triggered Pigeon with Travelling Salesman Problem
(DTP-TSP-D)
The node on the ground communicates with UAVs in the air to relay information. A
UAV serves as a transporting node in this routing system. The message is delivered
on time using a genetic algorithm. In terms of packet delivery ratio and average
latency, it beats the other conventional routing method.
2.2.9
Mobility Predication-Based Geographic Routing (MPGR)
Inter-UAV networks based on geographic location. MPGR detects the mobility of
UAVs using the Gaussian distribution function.
2.2.10
Geographic Position Mobility-Oriented Routing Protocol
(GPMOR)
Gauss–Markov mobility model was utilized to predict mobility routing. UAVs are
equipped with GPS and communicate their geographical location to nearby UAVs.
Additionally, GPMOR predicts and deduces the movements of neighbors.
182
2.2.11
A. S. Banu and G. Padmavathi
Optimized Link State Routing with Mobility and Delay
Prediction (OLSR-PMD)
OLSR-PMD uses the Kalman filter technique to find an inter relay of constant
neighbor nodes based on node movement and latency predictions.
2.2.12
Distance-Based Greedy Routing (DSGR)
DSGR seeks to speed up UAV route setting. It depends on local forwarding and avoids
route configuration. The network is grid-based. Each node’s location is measured.
DSGR outperforms Dijkstra’s shortest route.
2.2.13
Robust and Reliable Predictive Routing (RRPR)
RRPR merges both direct and indirect transmission by adjusting the angle. The
position and trajectory data are obtained using unicast and geocast routing methods.
2.2.14
Topology-Aware Routing Choosing Scheme (TARCS)
Topology modification is critical in FANETs. Moving nodes in TARCS detect periodic changes in network architecture. TARCS can adapt to FANET topology changes.
The new topology determines the routing route. Topology Change Degree is a
mobility measure used in FANETs to indicate topology change.
2.2.15
Aeronautical Mobile Ad Hoc Networks (ARPAM)
ARPAM packets include geographical information utilized in the routing process to
make the best choices depending on the nodes’ locations.
2.2.16
Reactive Greedy Reactive Protocol (RGR)
If there is no existing route to the desired location, the sender must create an ondemand route to maintain communication with the destination.
2.2.17
Geolocation-Based Multi-Hop Routing Protocol (GLMHRP)
After a while, UAV transmits route data. This data includes the UAV’s position,
speed, and direction. The data sending choice is based on greed. Navigation nodes
can receive the latest information about nodes, affecting routing performance.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
2.2.18
183
Position Aware, Secure, and Efficient Mesh Routing (PASER)
A complete safe routing route with valid nodes. It may rapidly identify malicious
nodes. PASER is better for dynamic UAV networks.
2.2.19
Location-Aided Delay Tolerant Routing (LADTR)
LADTR uses the Store-Carry-Forward (SCF) method to improve routing protocol
performance in UAV networks.
2.2.20
Jamming-Resilient Multipath Routing (JARMROUT)
Failures in limited areas or deliberate interferences and interruptions affect FANET’s
overall functionality. The JARMROUT is based on a mix of three central systems:
quality of the connection, the volume of traffic, and geographical distance.
2.2.21
Greedy Perimeter Stateless Routing (GPSR)
This technique selects a coordinator based on location. The GPSR protocol works by
passing data packets to the nearest neighbor node. Portion forwarding is used when
greedy forwarding fails.
2.2.22
Greedy-Hull-Greedy (GHG)
To restrict local recovery, this protocol splits the network into closed sub-spaces.
2.2.23
Greedy-Random-Greedy (GRG)
To find the local minimum, the message is forwarded using the greedy method.
However, weak networks do not fit this approach.
2.2.24
Greedy Forwarding (GF)
In contrast to the GRG, this routing algorithm only verifies the node position at every
step without making any indicators. After that, it sends the packet to the nearest node.
184
2.2.25
A. S. Banu and G. Padmavathi
Energy-Balanced Greedy Forwarding Routing (EBGR)
Partitioning the forwarding area into four sections. After this one in each potential
sub-region, the node is chosen based on its excess energy and the remaining hops to
the destination node.
2.2.26
Greedy Distributed Spanning Tree Routing (GDSTR)
Greedy forwarding maintains track of 2-hop neighbor information to avoid local
minima.
2.3 Hierarchical Routing
The lowest levels of hierarchical routing protocols may create clusters of nodes.
Each node has a database of information about its neighbors that is updated by hello
packets. To choose the optimal route, each cluster head interacts with the others.
Here is a collection of hierarchical routing protocols.
2.3.1
Cluster-Based Routing Protocol (CBRP)
Cluster-based networks may arrange UAVs. The CBRP is split into square grids
depending on geographic location. One UAV will serve as a Cluster Head, responsible
for data routing. Members of every UAV cluster send data to the cluster head for base
station transfer. It preserves routing tables and does not need to find routes, which
reduces overhead.
2.3.2
Modularity-Based Dynamic Clustering Relay Routing Protocol
(MDCR)
After creating clusters that change throughout time, The UAVs are relocated to
vertically projected positions from the cluster centroids. Modularity is assessed and
compared inside and across clusters of a network graph. Modularity is assessed and
compared inside and across clusters of a network graph. Some of the advantages are
transmission power and energy efficiency.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
2.3.3
185
Bio-inspired Clustering Scheme for Fanets (BICSF)
The BICSF method utilizes energy-aware cluster creation and cluster head selection. BICSF has performed better than others with cluster construction time, energy
consumption, cluster lifespan, and delivery success probability.
2.3.4
Hybrid Self-organized Clustering Scheme (HSCS)
It’s a hybrid clustering algorithm of dragonfly algorithm and glowworm swarm optimization used mainly in cognitive IoT-based UAV networks. The HSCS system uses
GSO to create clusters and choose cluster heads and DA for efficient cluster management. A method to detect deceased cluster members is used to improve network
stability. The route selection function in HSCS selects the next-hop neighbor for
data transfer.
2.3.5
Swarm Intelligence-Based Localization and Clustering (SIL-SIC)
The SIL algorithm leverages particle fitness function for intercluster distance, intracluster distance, residual energy, and geographic location with PSO. SIC saves
energy, and better particle optimization selects cluster heads. The SIL method
improves convergence time and accuracy while reducing computing costs.
2.3.6
Cluster-Based Routing Protocol (CBRP)
To arrange the nodes into overlapping or disjoint clusters, this routing technique
uses each cluster head and membership information to identify inter-cluster routes.
In order to speed up the route discovery process, the protocol clusters nodes into
groups. Inter-cluster and intra-cluster routing use unidirectional connections.
2.3.7
Enhanced Cluster Head Gateway Switch Routing (ECGSR)
Here AODV-based routing method has a congestion management mechanism. The
cluster head observes traffic jams by decreasing transmissions and establishes pathways as needed. This method minimizes packet drops, routing overhead, and delay
while decreasing network congestion.
186
2.3.8
A. S. Banu and G. Padmavathi
Fisheye State Routing (FSR)
Updates a topological map used to calculate the shortest route. FSR is a protocol that
utilizes link-state information to route traffic that performs three functions: Neighbor
find Route computation, information distribution.
2.4 Probabilistic Routing Protocols
For network congestion and security, probabilistic routing systems discover
numerous paths from source to destination. Here are some examples of UAV network
algorithms.
2.4.1
Random Walk Routing (RWR)
A “random walk” method forwards packets to the next available neighbor. Attackers
cannot anticipate the route, which offers excellent security. However, forwarding is
inefficient.
2.4.2
MIMO-Based Random Walk Routing (MRWR)
This routing system adapts communication modes at each step to save energy over
random walk routing. The greater complexity costs more.
2.5 AI-Enabled Routing Protocols
AI-powered routing protocols utilize ML algorithms to understand the architecture of
the network, status of the channel, user behavior, the movement of traffic, and other
factors. Using these methods, current networking may be created, especially for
dynamic UAV networks. The following is a comprehensive classification of artificial
intelligence protocols.
2.5.1
Topology Predictive Routing Protocols
Machine Learning algorithms are used to predict node motion trajectories in order to
choose routes. The edge measurements may include distance, energy consumption,
latency, bitrate, etc. Topology-based routing protocols include.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
187
Learning-Based Adaptive Position MAC Protocol (LAP-MAC)
Here protocol combines the concept of PPMAC and RLSRP and performs better in
terms of directional deafness.
Predictive Dijkstra’s
It implies that the intermediary nodes’ positions are predicted using ML techniques.
The shortest route based on Dijkstra’s is found by combining predicted information
with path specifications.
Predictive Greedy Routing (PGR)
To get to the target node, each node guesses its neighbors’ whereabouts.
Predictive Optimized Link State Routing (P-OLSR)
This method uses GPS data to compute an Expected Transmission (ETX) count
measure to assess the nature of the connection while determining the optimum route.
Geographic Position Mobility Oriented Routing (GP-MOR)
To reduce the effect of highly dynamic UAV motions, it adopts the Gauss Markov
mobility model. This method uses the mobility connection as well as the Euclidean
distance for better choices.
Mobility Prediction Clustering Algorithm (MPCA)
Helps find cluster heads in clustered environments based on node reliability and
predicts the network topology. This method also guarantees cluster stability.
Robust and Reliable Predictive (RARP)
This technique uses a combination of unicasting and geocasting routing technologies
to route traffic. Directional transmission is used to monitor topology changes and
guess intermediate node positions using a 3-D estimate.
188
A. S. Banu and G. Padmavathi
Scoped Flooding and Mobility Prediction-Based RGR (SFMPRGR)
A distance between two nodes is computed by combining data packets’ mobility
prediction (velocity, direction, timestamp). Since GGF saves lost data packets if the
following hop is out of range, this method works well for dynamic networks.
Q-Learning-Based Geographic Ad Hoc Routing Protocol (QGeo)
Nodes make geographic routing choices dynamically with no prior knowledge of
the network’s architecture. It has a neighbor table and a Q-learning element. The
GPS’s position estimate module continually updates the GPS’s stated location or
other localization techniques’ current location.
Predictive Ad Hoc Routing Fueled by Reinforcement Learning
and Trajectory Knowledge (PARRoT)
As with the previous routing protocol, this one-use mobility control information
predicts future agent movement, predicts its future location based on its present
position, and propagates it to neighboring nodes.
Fuzzy Logic Reinforcement Learning-Based Routing Algorithm (FLRLR)
Fuzzy logic is applied to find a live view of the closest nodes. It then uses reinforcement learning to decrease the typical hops produced via training and receive future
benefits.
2.5.2
Self-adaptive Learning-Based Routing Protocols
Most learning-based routing systems utilize Reinforcement Learning (RL) to learn
how to route. In addition to being independent of topologies modeling and channel
estimation, RL-based algorithms provide several other advantages. These learningbased routing algorithms have been developed to support dynamic UAV networks
better, as seen below.
Q-Routing
It learns based on experience. They are stored in a Q-table by each node. Each node
chooses the following node that reduces trip time. The preceding node’s Q-values
are updated after receiving a packet.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
189
Predictive Q-Routing (PQ-Routing)
By predicting traffic trends, they learned and stored new optimum policies under
decreasing load circumstances. To re-examine the routes that have been abandoned
owing to traffic delays. To modify the probing frequency, use the route recovery rate
estimate.
Dual Reinforcement Q-Routing (DRQ-Routing)
Each communication hop is based on the transmitter and recipient of each
communication hop adding data to the packet they get from their neighbors.
Credence-Based Q-Routing (CrQ-Routing) and Probabilistic Credence-Based
Q-Routing (PCrQ-Routing)
With these two techniques, congestion traffic is captured to enhance learning process.
Full-Echo Q-Routing
Each node gets the expected journey time of each neighbor, which is used to update
each neighbor’s Q-values.
Full-Echo Q-Routing with Adaptive Learning Rate
Adaptive learning is combined with full-echo Q-Routing to boost exploration
efficiency.
Adaptive Q-Routing with Random Echo and Route Memory (AQRERM)
the baseline method’s overshoot and settling time, and learning stability have been
improved.
Poisson’s Probability-Based Q-Routing (PBQ-Routing)
For intermittently linked networks, this method utilizes transmission probability and
Poisson’s probability to manage transmission energy.
190
A. S. Banu and G. Padmavathi
Delayed Q-Routing (DQ-Routing)
The value function is not overestimated, so the routing protocol is employed with
two sets of value functions.
QoS-Aware Q-Routing (Q2-Routing)
It guarantees traffic Quality of Service by varying the learning rate depending on
Q-value changes.
Q-Network Enhanced Geographic Ad Hoc Routing Protocol Based on GPSR
(QNGPSR)
Q-network is used to assess the quality of various route pathways. Q values decide
the forwarding. This data is used to evaluate the environment and node categories.
Adaptive and Reliable Routing Protocol with Deep Reinforcement Learning
(ARdeep)
Automatically characterizes network changes using a Markov Decision Process
model for formulating routing decisions considering node’s remaining energy, the
distance between the nodes, link status, the link’s expected connection time, the
packet error ratio.
Traffic-Aware Q-Network Enhanced Routing Protocol Based on GPSR
(TQNGPSR)
As a traffic balancing technique, this algorithm utilizes neighbor congestion information to assess the quality of a wireless connection. The protocol then decides on
routing depending on each wireless link’s assessment.
Q-learning Based Multi-objective Optimization Routing Protocol (QMR)
This new method adapts the learning parameters to the network’s dynamics.
Unknown optimum routes are investigated while re-estimating adjacent connections
to choose the most trustworthy next hop.
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
191
Q-Learning-Based Fuzzy Logic for Multi-objective Routing Algorithm
in Flying Ad Hoc Networks (QLFLMOR)
This method aids in selecting routing routes based on per-link and total path performance. Each UAV determines the best route to the target using a fuzzy system with
the link- and path-level characteristics.
Adaptive UAV-Assisted Geographic Routing with Q-Learning (QAGR)
When a routing request comes in, the ground vehicle looks up the Q-table filtered
according to the global routing route to see which node is the best fit.
Fully-Echoed Q-Routing with Simulated Annealing Inference for Flying Ad
Hoc Networks (FESAIQ-Routing)
The temperature parameter quantifies the effect of node mobility on Q-value
update rates. The gradual change in the exploration step improves exploration and
accommodates sudden changes in network dynamicity.
2.6 Deterministic Routing Protocol
A node’s subsequent motion is known to its neighbors. Because UAVs fly in regulated
formations, this protocol may be used in networks. A tree method might be used to
choose routes if all nodes are aware of the node movement, accessibility, and behavior
of other nodes. Nodes are classified as child or root nodes in a tree. The shortest path
is chosen from the tree.
2.7 Stochastic Routing Protocols
The networks that exhibit unexpected behavior make use of this protocol. In this
case, packet delivery choice is critical. This may be done by re-sending the data to
the next node. Some of the stochastic routing protocols include.
•
•
•
•
Epidemic Routing-based Approach
Estimation-based Routing
Node Movement and Control-based Routing, and
Coding-based Routing.
192
2.7.1
A. S. Banu and G. Padmavathi
Epidemic Routing-Based Approach
Moving nodes are not linked with each other. The other nodes are flooded with
the same messages several times. Nodes estimate the likelihood of each connection
instead of sending messages to the following nodes. The node needs big buffers,
bandwidth, and power for this approach.
2.7.2
Estimation-Based Approach
Each node stores the packet and forwards it depending on the estimate. Small
networks benefit from random nodes, while extensive networks suffer from estimated
overheads.
2.7.3
Node Movement and Control-Based Approach
When detached from adjacent nodes, and nodes determine whether to wait for reconnection. In a reactive situation, waiting for re-connection may cause unacceptable
transmission delays.
2.7.4
Coding-Based Approach
This protocol utilizes network coding to minimize duplicating data and retransmission. This technique may be used in UAV networks when retransmission requires
finding a new route due to interruption.
2.8 Social Network-Based Approach
Using, a wide variety of networking protocols is unrealistic when nodes’ mobility is
fixed. When nodes visit a location, they may save the location’s data in a database.
Using this information, the node may rapidly pick routes for future tries. This protocol
is helpful for UAV node information storage. SN-based routing demands more buffer
space and bandwidth.
3 Vulnerabilities in UAV
Due to the high usage of UAV networks in smart cities, Security, Safety, and Privacy
have become the primary concern. The main task of the UAVs is to gather information, analyze, and transmit sensitive data. Attackers try to impersonate or poison the
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
193
Fig. 4 Different types of vulnerabilities based on GNSS
routing table in the routing protocols to perform the attack. So secured communication becomes a significant concern because UAVs are vulnerable to cyber-attacks,
data manipulation, and data theft [1].
The effects vary as a result of the different tactics and targets of attacks. Modern
UAVs depend largely on Global Navigation Satellite System (GNSS) for guidance,
navigation, and control (GNC). The Global Positioning System (GPS) is the most
commonly used GNSS. GPS-dependent UAVs need precise, reliable, and continuous
location data to operate safely. However, research has revealed that GPS signals may
be jammed or spoofed due to intrinsic flaws. The civil GPS systems lack encryption
and authentication, making the satellite signals readily replicable or fabricated to
launch GPS spoofing attacks. Numerous vulnerabilities exist with GNSS, which is a
synonym of GPS given in Fig. 4. According to the Royal Academy of Engineering
[10], GNSS vulnerabilities fall across three main categories:
3.1 System-Related Vulnerabilities
The satellites themselves (for example, a lack of available satellites or the transmission of poor signals); receiver failure and outage; or issues with the augmentation
systems that boost the GNSS signal (e.g. EGNOS). This covers both failures induced
by assaults on GNSS systems and outages caused by the orbital environment, such
as geomagnetic storms, among other things.
194
A. S. Banu and G. Padmavathi
3.2 Propagation Channel Vulnerabilities
It includes signal interruptions caused by environmental variations caused by space
weather phenomena and mistakes caused by signals reflected off buildings or other
objects (multipath).
3.3 Interference’s Vulnerabilities
They are the disrupted signals from other sources. Commercial power transmitters
or radars may create unintentional interference; however, it is also possible that
receivers will be jammed on purpose, sending false GPS signals (GPS Spoofing) and
delayed or rebroadcasted signals (meaconing).
Particular vulnerabilities based on interferences attempt to steal data through security flaws in communication connections, while others attempt to spoof sensors, such
as GPS spoofing and jamming. As with conventional network security, the Availability, Confidentiality, Integrity of UAV communications are all security objectives.
GPS Spoofing and Jamming attacks compromise the network, which challenges the
CIA rule Fig. 5 and tries to hijack the UAV, intercept the communication, and act
as MTM attacks. Additional characteristics like privacy, authenticity, accountability,
non-repudiation, and reliability may be the goal of UAV security, depending on the
specific prerequisites. GPS Spoofing and Jamming attacks are discussed in detail in
the coming sections.
Fig. 5 Objectives of security
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
195
Fig. 6 GPS spoofing attack
4 GPS Spoofing and Jamming Attacks
4.1 GPS Spoofing Attack
GPS Spoofing is duplicating or falsifying GPS signals to mislead a targeted GPS
device or recipient, altering its characteristics such as position, velocity, and time. If
a GPS spoofing attack is successful, it might cause a drone to crash or alter the flying
path. GPS Spoofing is explained in Fig. 6. Various studies indicate that an attacker
may push a GPS-guided drone off course or even hijack it if they know its present
location and planned flight route. A drone may be made to fly across no-fly zones by
spoofing the “Geo-fencing” safety function. This weakness allows drug traffickers
and others to cross prison boundaries for drug trafficking and monitoring illegally.
Spoofing is more subtle: a fake signal sent by a ground station that merely confuses
a satellite receiver [11].
4.2 Jamming Attack
Jamming is typically caused by GNSS signal interference. However, accidental
jamming may occur due to space weather or defective equipment that emits signals
on the L1 frequency, interfering with GNSS reception. Intentional jamming is used to
overwhelm weak GNSS receivers [12]. Personal Protection Devices (PPD) are often
employed in addition to military jammers. These are easily accessible and cheap,
yet most nations ban them. GPS has an inherent vulnerability that has existed since
196
A. S. Banu and G. Padmavathi
Fig. 7 Jamming attack
the system’s inception in the 1980s. This is an unavoidable problem since GPS is
intended to be shared and utilized by all civilian devices, exposing the signal shown
in Fig. 7.
Table 1 details many real-world instances of GNS spoofing and jamming attacks
against UAVs [13].
Cyber security designs are classified into two broad categories: cyber defense and
cyber detection. On the one hand, the former method focuses on data privacy and
confidentiality and ways to mitigate outsider attacks (i.e., Attackers located outside
the local network’s borders) that are most inclined to compromise network integrity.
In contrast to that, later methods are mostly employed to identify network intrusions.
As a result, they are capable of detecting internal and external attacks with incredible
accuracy. These latter methods depend on Intrusion Prevention Systems (IPS) and
Intrusion Detection Systems (IDS) to anticipate and identify an attacker’s misbehavior, respectively. With the continuous growth in research of UAVs and their security and privacy issues, this chapter comprises various types of Machine Learning,
Deep Learning, Intrusion Detection System, and Intrusion Prevention System models
for detection and prevention of GPS Spoofing and Jamming attacks are covered in
Sect. 5. As mentioned above, some countermeasures to the attacks are covered in
Sect. 3 and conclude with the future prediction of GPS Spoofing and Jamming attacks.
5 Literature Survey
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
197
Table 1 GPS spoofing and jamming attacks real-time incidents
Month, year
Actual incidents of GPS spoofing and jamming UAV attacks
Dec 2019
This periodic GPS signal loss has been linked to a jammer that was set up at a
pig farm near Harbin Airport in northeast China
Jan 2020
After discovering that GPS jamming devices are utilized in 85% of the
country’s cargo truck thefts, Mexico adopts anti-jammer legislation
Feb 2020
A light aircraft pilot’s worrisome report to NASA’s Aviation Safety Reporting
System indicates that a US Department of Defense (DoD) drone may be faking
the signals
A GPS and Galileo signal interruption has been reported at a French GNSS
equipment manufacturer’s facility regularly
Mar 2020
‘Circle-style’ In Iran’s capital, Tehran, GPS spoofing has been recorded. An
unidentified GPS user alerts the US government that his or her (unspecified)
gadget seems to be moving in a circle around the Iranian Army training
institution while it is parked
Jun 2020
As in the past, GPS jamming is causing havoc in the extreme north of Norway,
not far from the Russian border
Aug 2020
After a drone accident in the UK, the dangers of jamming and spoofing
interference to unmanned aerial vehicles have been brought to light (UAVs)
Many Chinese fishing boats have been accused of lying about their
whereabouts to conceal illegal fishing operations
Sep 2020
The Maritime Administration of the United States urges the maritime sector to
be alert to GPS interruptions anywhere in the world
Nov 2020
According to Fortune, GPS failures are now frequent on commercial aircraft
routes between the United States, Europe, and the Middle East, echoing the
MARAD warning from September
Aug 2021
Iran-Israel maritime tensions rise as a deadly drone strike targets a ship
Sep 2021
GPS Spoofing in Pokemon Go
Author
Techniques used
Enhancements
Zouhri et al. [14]
Security Communication
Protocol Between UAV and
GCS (SPUAV)
Confidentiality, integrity,
authentication based on sufficient
energy, network connectivity,
mobility, and network security
Chen et al. [15]
Lightweight active GPS
spoofing detection method
Real-time detection without
reducing the detection rate
Haque et al. [16]
Identity-based encryption and Security and confidentiality
selective encryption methods flexibility, efficiency,
communication, storage, overhead
reduction, and data hiding
mechanism
Wang et al. [17]
LSTM model
Detects in a short time and
quickly, enhances computing
efficiency, no upgradations
needed, tested in various platforms
(continued)
198
A. S. Banu and G. Padmavathi
(continued)
Author
Techniques used
Enhancements
Dasgupta et al. [18]
LSTM model for spoofing
detection
The maximized attack detection
rate
Jansen et al. [19]
Crowd-GPS-Sec
Detects the attack in <2 s and
localizing spoofing devices
Yu et al. [20]
SNORT IDS
Detects several attacks at once
Feng et al. [21]
Two-step GA-XGBoost
method
Detects an attack in a second
consuming time
Varshosaz et al. [22]
Visual Odometry (VO)-vision Efficient and effective
based spoofing detection
method
Zou et al. [23]
UAV model estimate-based
detection model
Eldosouky et al. [24]
Mathematical approach based Reduces the possibility of the
on system dynamics
attack
Semal et al. [25]
Certificateless-group
authenticated key agreement
(CL-GAKA) scheme
Resilient against several attacks,
enhance Confidentiality, Integrity,
and authenticity
Liang et al. [26]
Time difference-based
positioning scheme
Improve performance without
reducing detection rate
Arthur [27]
Self-taught learning (STL)
IDS with multi class SVM
Lightweight IDS consumes low
energy, produce high actual
positive values
Qiao et al. [28]
Lucas-Kanade (LK) method
Detects the attack in 5 s, more
effective
Manesh et al. [29]
Artificial neural network
Low false alarm rate
Calvo-Palomino et al. [30]
LSTM based detection
method
Accurate prediction of the attack
Xiao et al. [31]
Recurrent neural networks
Detects abnormal behavior nodes
in time with high accuracy
Sedjelmaci et al. [32]
Rules-based IDS
Efficient, lightweight, also detect
false information dissemination
attack, black hole attack and grey
hole attack, low communication
overhead, less false-positive rates,
and high detection accuracy
Meng et al. [33]
Linear Regression (LR) based Practical, easy to implement
anti-spoofing model
Xue et al. [34]
Deep SIM, a satellite imagery The high detection rate in
matching approach
<100 ms require no alterations in
existing GPS
Diao et al. [35]
Signal noise ratio
High tracking performance on
time
Performed well with 98%
detection accuracy
(continued)
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
199
(continued)
Author
Techniques used
Enhancements
Whelan et al. [36]
One class classifier IDS
Compatible with a variety of UAV
platforms, favorable outcomes
Meng et al. [37]
Drone Sensor Spoofing
Detection (SSDGOF)
algorithm
Immediate detection of spoofing
but lacks in finding out the
location of the spoofer
Titouna et al. [38]
Bayesian network
Reliable, schedule maintenance
on time, high security
Bouhamed et al. [39]
IDS based on Deep Q
Learning (DQN)
The offline learning approach is
made periodically; high efficiency
consumes energy
It is observed based on the literature that, unlike cyber-attacks, GPS spoofing
and jamming assaults are effective against control systems that include protected
software, data, and communications networks, making them a longer-term threat.
Traditional communication security methods such as anomaly detection procedures
cannot effectively secure real-time data transfer for UAVs since they frequently
utilize customized protocols and transmission technologies. Numerous writers are
attempting to provide methods for detecting GPS signals. Spoofing and jamming
attacks with a high detection rate, low energy consumption, and sometimes acting as
IDPs, identified a wide variety of different kinds of attacks using a learning method.
However, literature focusing on or combining security and other issues such as
increasing payload energy usage and air traffic management have not been studied,
and expected that hybridization of deep learning and artificial intelligence algorithms
could provide promising results soon. As a result, in order to develop distributed
security policies and methods with high usability and security efficacy for UAV ad
hoc networks, the study is required. On the other hand, network security is seen as
a method to guarantee the security of regular network operations; vast chunks of
resources on nodes are not permitted to be occupied. As a result, tighter security
measures should not degrade network performance or interfere with the network’s
normal functioning. As a result, future research must strike a balance between the
strength of security measures and the performance of the network connection.
6 Conclusion
UAVs are rapidly being utilized for military and civilian purposes. Communication security is a critical element in the proper functioning of UAVs. Otherwise,
attackers may seize UAVs by launching GPS Spoofing and Jamming attacks. Unauthenticated GPS signals may divert, threaten, damage, or even hijack a faked UAV.
Security experts have been drawn to GPS because of its ubiquitous use and inherent
vulnerabilities. Making highly effective detection systems with low false alarms and
miss-detection rates is challenging. Existing detection techniques are minimal. One
200
A. S. Banu and G. Padmavathi
issue is that most suggested solutions involve changing UAV network architecture
and communication standards. Also, most of these techniques have low detection
rates and significant false alarm rates.
References
1. D. He, S. Chan, M. Guizani, Communication security of unmanned aerial vehicles. IEEE Wirel.
Commun. 24(4), 134–139 (2017). https://doi.org/10.1109/MWC.2016.1600073WC
2. https://www.shapingtomorrow.com/home/alert/6808171-Drone-Update
3. H. Sedjelmaci, S.M. Senouci, M. Messous, How to detect cyber-attacks in unmanned aerial
vehicles network?, in 2016 IEEE Global Communications Conference (GLOBECOM) (2016),
pp. 1–6. https://doi.org/10.1109/GLOCOM.2016.7841878
4. https://dronedj.com/2019/11/25/drone-industry-100-billion-2020/
5. https://www.auav.com.au/articles/drone-types/
6. M.Y. Arafat, S. Moh, Routing protocols for unmanned aerial vehicle networks: a survey. IEEE
Access 7, 99694–99720 (2019). https://doi.org/10.1109/ACCESS.2019.2930813
7. Q. Sang, H. Wu, X. Ling, P. Xie, Review and comparison of emerging routing protocols in
flying ad hoc networks. Symmetry 12(6), 971 (2020)
8. A. Rovira-Sugranes, F. Afghah, J. Chakareski, A. Razi, A review of AI-enabled routing protocols for UAV networks: trends, challenges, and future outlook. arXiv preprint arXiv:2104.
01283 (2021)
9. J. Jiang, G. Han, Routing protocols for unmanned aerial vehicles. IEEE Commun. Mag. 56(1),
58–63 (2018)
10. The Royal Academy of Engineering, Global navigation space systems: Reliance and
vulnerabilities (The Royal Academy of Engineering, London, 2011)
11. S.Z. Khan et al., On GPS spoofing of aerial platforms: a review of threats, challenges, methodologies, and future research directions. PeerJ Comput. Sci. 7, e507 (2021). https://doi.org/10.
7717/peerj-cs.507
12. https://www.crfs.com/blog/how-to-deal-with-gps-jamming-and-spoofing/
13. D.A. Goward, Thousands of GNSS Jamming and Spoofing Incidents Reported (2020)
14. O. Zouhri, S. Benhadou, H. Medromi, A new adaptative security protocol for UAV network, in
International Symposium on Ubiquitous Networking (Springer, Singapore, 2016), pp. 649–657
15. X. Chen, X. Huang, J. Zhang, Machine Learning for Cyber Security: Second International
Conference, ML4CS 2019, Xi’an, China, September 19–21, 2019, Proceedings, vol. 11806
(Springer Nature, 2019)
16. M.S. Haque, M.U. Chowdhury, A new cyber security framework towards secure data communication for unmanned aerial vehicle (UAV), in International Conference on Security and Privacy
in Communication Systems (Springer, Cham, 2017), pp. 113–122
17. S. Wang, J. Wang, C. Su, X. Ma, Intelligent detection algorithm against UAVs’ GPS spoofing
attack, in 2020 IEEE 26th International Conference on Parallel and Distributed Systems
(ICPADS) (2020), pp. 382–389. https://doi.org/10.1109/ICPADS51040.2020.00058
18. S. Dasgupta, M. Rahman, M. Islam, M. Chowdhury, Prediction-based GNSS spoofing attack
detection for autonomous vehicles, in Transportation Research Board 100th Annual Meeting,
Transportation Research Board (Washington, DC, USA, 2021)
19. K. Jansen, M. Schäfer, D. Moser, V. Lenders, C. Pöpper, J. Schmitt, Crowd-gps-sec: leveraging
crowdsourcing to detect and localize gps spoofing attacks, in 2018 IEEE Symposium on Security
and Privacy (SP) (IEEE, 2018), pp. 1018–1031
20. J. Yu, B.-M. Cho, K.-J. Park, Detecting UAV attacks through network monitoring, in Vertical
Flight Society (2019)
21. Z. Feng, N. Guan, M. Lv, W. Liu, Q. Deng, X. Liu, W. Yi, Efficient drone hijacking detection
using two-step GA-XGBoost. J. Syst. Architect. 103, 101694 (2020)
Taxonomy of UAVs GPS Spoofing and Jamming Attack …
201
22. M. Varshosaz, A. Afary, B. Mojaradi, M. Saadatseresht, E.G. Parmehr, Spoofing detection of
civilian UAVs using visual odometry. ISPRS Int. J. Geo-Inform. 9(1), 6 (2020)
23. Q. Zou, S. Huang, F. Lin, M. Cong, Detection of gps spoofing based on UAV model estimation,
in IECON 2016—42nd Annual Conference of the IEEE Industrial Electronics Society (IEEE,
2016), pp. 6097–6102
24. A.R. Eldosouky, A. Ferdowsi, W. Saad, Drones in distress: a game-theoretic countermeasure
for protecting UAVs against gps spoofing. IEEE Internet Things J. 7(4), 2840–2854 (2019)
25. B. Semal, K. Markantonakis, R.N. Akram, A certificateless group authenticated key agreement
protocol for secure communication in untrusted UAV networks, in 2018 IEEE/AIAA 37th
Digital Avionics Systems Conference (DASC) (IEEE, 2018), pp. 1–8
26. C. Liang, M. Miao, J. Ma, H. Yan, Q. Zhang, X. Li, Detection of global positioning system
spoofing attack on unmanned aerial vehicle system. Concurr. Comput. Pract. Exp., e5925
(2020)
27. M.P. Arthur, Detecting signal spoofing and jamming attacks in UAV networks using a
lightweight IDS, in 2019 International Conference on Computer, Information and Telecommunication Systems (CITS) (IEEE, 2019), pp. 1–5
28. Y. Qiao, Y. Zhang, X. Du, A vision-based GPS-spoofing detection method for small UAVs, in
2017 13th International Conference on Computational Intelligence and Security (CIS) (IEEE,
2017), pp. 312–316
29. M.R. Manesh, J. Kenney, W.C. Hu, V.K. Devabhaktuni, N. Kaabouch, Detection of GPS
spoofing attacks on unmanned aerial systems, in 2019 16th IEEE Annual Consumer Communications and Networking Conference (CCNC) (IEEE, 2019), pp. 1–6
30. R. Calvo-Palomino, A. Bhattacharya, G. Bovet, D. Giustiniano, Short: LSTM-based GNSS
spoofing detection using low-cost spectrum sensors, in 2020 IEEE 21st International Symposium on “A World of Wireless, Mobile and Multimedia Networks” (WoWMoM) (IEEE, 2020),
pp. 273–276
31. X. Ke, J. Zhao, Y. He, C. Li, W. Cheng, Abnormal behavior detection scheme of UAV using
recurrent neural networks. IEEE Access 7, 110293–110305 (2019)
32. H. Sedjelmaci, S.M. Senouci, N. Ansari, A hierarchical detection and response system to
enhance security against lethal cyber-attacks in UAV networks. IEEE Trans. Syst. Man
Cybernet. Syst. 48(9), 1594–1606 (2017)
33. L. Meng, L. Yang, S. Ren, G. Tang, L. Zhang, F. Yang, W. Yang, An approach of linear
regression-based UAV GPS spoofing detection, in Wireless Communications and Mobile
Computing (2021)
34. N. Xue, L. Niu, X. Hong, Z. Li, L. Hoffaeller, C. Pöpper, DeepSIM: GPS spoofing detection on
UAVs using satellite imagery matching, in Annual Computer Security Applications Conference
(2020), pp. 304–319
35. D. Diao, X. Wang, C. Yang, Y. Wang, J. Ma, GPS spoofing detection via SNR, in Proceedings
of the 1st ACM International Workshop on Security and Safety for Intelligent Cyber-Physical
Systems (2020), pp. 1–5
36. J. Whelan, T. Sangarapillai, O. Minawi, A. Almehmadi, K. El-Khatib, Novelty-based intrusion
detection of sensor attacks on unmanned aerial vehicles, in Proceedings of the 16th ACM
Symposium on QoS and Security for Wireless and Mobile Networks (2020), pp. 23–28
37. L. Meng, S. Ren, G. Tang, C. Yang, W. Yang, UAV sensor spoofing detection algorithm based
on gps and optical flow fusion, in Proceedings of the 2020 4th International Conference on
Cryptography, Security and Privacy (2020), pp. 146–151
38. C. Titouna, F. Naït-Abdesselam, A lightweight security technique for unmanned aerial vehicles
against GPS spoofing attack, in 2021 International Wireless Communications and Mobile
Computing (IWCMC) (2021), pp. 819–824. https://doi.org/10.1109/IWCMC51323.2021.949
8734
39. O. Bouhamed, O. Bouachir, M. Aloqaily, I.A. Ridhawi, Lightweight IDS for UAV networks:
a periodic deep reinforcement learning-based approach, in 2021 IFIP/IEEE International
Symposium on Integrated Network Management (IM) (2021), pp. 1032–1037
Investigation on Challenges of Big Data
Analytics in UAV Surveillance
N. Vanitha , G. Padmavathi , P. Nivedha, and K. Bhuvana
Abstract In today’s world, there is a tremendous need for UAV surveillance to
maintain safety and security purposes. The abbreviation for UAV is an Unmanned
Aerial Vehicle, which is frequently known as drones. A drone is an airship without
any human being pilot. UAVs are a component of unmanned airship systems (UAS),
that includes an additional ground-based controller. Initially, UAV has developed the
De Havilland DH. 82B Queen Bee airship, which is applied for a low-price radiocontrolled drone. This chapter mainly focuses on applications of drones, the significance of big data in UAV surveillance, challenges of big data in UAV surveillance,
conclusion, and Future work of UAV surveillance.
Keywords Unmanned aerial vehicle · Surveillance · Bigdata analytics ·
Significance · Machine learning
1 Introduction
1.1 UAV Surveillance
In the modern era, drones are also known as Unnamed Aerial Vehicle (UAV), which is
considered one of the most emerging technologies. UAVs are motorized area vehicles
that can be remotely accessed by a combination of capabilities. A drone is a flying
robot that will be remotely controlled by software that controls the flight plane’s name
water system with the help of onboard sensors and GPS. From aerospace industries
to monitoring crops, drone features of providing to be highly beneficial in every
place were unable to perform inappropriately timely and effective manner. Increasing
N. Vanitha (B) · P. Nivedha · K. Bhuvana
Dr. N.G.P. Arts and Science College, Coimbatore, India
e-mail: vanitha@drngpasc.ac.in
N. Vanitha · G. Padmavathi
Avinashilingam Institute for Home Science and Higher Education for Women, Coimbatore, India
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_11
203
204
N. Vanitha et al.
Fig. 1 Unmanned aerial vehicle in surveillance
work efficiency and productivity are the few of the top users of drone technology of
industries. Figure 1 shows the Unmanned aerial vehicle in surveillance.
1.2 Application of UAV Surveillance
Some of the applications of UAV surveillance includes the following,
•
•
•
•
•
Aerial photography in journalism and films,
Speedy shipping and delivery,
Road Safety,
Forest fire detection and
Disaster management.
1.2.1
Aerial Photography in Journalism and Films
Creative and artistic fields like cinemas and Television are continuously searching
for extraordinary images. Their main aim is to convey their information most
appropriately [1, 2].
1.2.2
Speedy Shipping and Delivery
A delivery drone is used to deliver packages, Medical supplies, food, and other goods.
They are currently being used in a few locations around the United States of America.
Investigation on Challenges of Big Data Analytics …
1.2.3
205
Road Safety
The application of UAV in road safety is tremendous. It includes Accidental
Investigation, Risk Assessment and overall surveillance of road [3].
1.2.4
Forest Fire Detection
Real-time controlling and monitoring forest fire is very difficult in protecting the
forest. By using UAV in forest fire detection, we can find the fire in a forest that can
significantly shorten the reaction time and reduce the potential damage and damage
cost [4, 5].
1.2.5
Disaster Management
During a natural disaster like Tsunami, a quick drone can survey the affected area
and send key factors about the damage in a particular area [6, 7].
1.3 Importance of UAV Surveillance
The cost of the drone program is less than other traditional methods of surveying.
Drones are used for security purposes which have the ability to reach difficult locations more quickly. Drones can be used in precision agriculture for a variety of
farming needs. Drones are used to gather valuable data during and after a natural
disaster to aid in security. It is very cheaper to produce than planes. UAVs are used
in many intervals because of their safety purpose. With the help of their remotecontrolling capability, drones can observe places, transmit through impossible risk,
and they can even notify intimidate conditions such as cleanser. Not only this but
Drone Technology can also be employed in the armed during difficult periods as
well.
2 Big Data Analytics
Big data is an immeasurable pasture that handles how to inspect and consistently take
out the sectional information from the data sets that are too large or too complex to
solve. By using customary data-storing application software, Big data analytics has a
vast dare to abduction the data, storing data, analyzing the data, searching, share the
cost, convey, evoke, interrogation, renovate, information privacy, and data source.
206
N. Vanitha et al.
2.1 Importance of Big Data Analytics
The big data analytics field assists organizations tackle their data and to employ and
spot new occasions. This cause intelligent business moves, and other efficient operations, higher-level profits, and satisfying customer’s needs. Big Data helps companies
to give risk to valuable insights. Many companies are using Big Data to clarify their
marketing effort and techniques. We can’t bracket big data to any certain data volume.
Organizations are using big data analytics systems and software to build data-driven
resolutions which can improve business-related outcomes. It includes many benefits like powerful marketing, new takings opportunities, regular actualization, and
improved operational efficiency. With this effective strategy method, these benefits
can provide a competitive edge over the opponent.
2.2 Significance of Big Data Analytics in UAV Surveillance
Drones can gather data about the release, land surveying, and wireless telegraph.
They have the potential that helps in an assortment of industries. While drones are
valued for the images and videos they collect, they are increasingly able to store other
types of information, including radio signals and soil moisture. The amount of data
that they gather is huge. Here, Big data play a vital role in collecting information.
Natural disaster management is a good example that they can gather information and
analyze a vast amount of data, categorize it, store it, and develop it.
3 Background Study
The primary objective of the investigation is to explore the challenges of the Bigdata
Analytics in UAV Surveillance.
The following Table 1 illustrates the literature study of UAV Surveillance in Big
Data Analytics.
Khana et al. [8] investigated the Flight Planning period includes the preparation
for the implement of a perfect UAV flight for collecting exact data. With an increase
in the notable number of UAVs the laws are now being implemented all over the
world. In this type of situation, the UAV flight are planning to step forward and
become much more critical [8].
Hildmann et al. [9] investigated the using of flying video surveillance systems. It
is popular visions in the forthcoming culture, so they can be flying cellular networks
that are used to gather extreme attention in searching community. Ultimately working
on the wired infrastructure constructed network. UAV are relatively independent at
fixed entry points [9].
Investigation on Challenges of Big Data Analytics …
207
Table 1 Background study
Year
Author name
2016
Muhammad Arsalan Khana,
UAV-based traffic analysis: a
Wim Ectorsa, Tom Bellemansa, universal guiding framework
Davy Janssensa and Geert Wets based on literature survey
Journal name
Technique used
2019
Hanno Hildmann, and Ernö
Kovacs
Using unmanned aerial
Mobile aerial
vehicles (UAVs) as mobile
communication
sensing platforms (MSPs) for infrastructure
disaster response, civil security
and public safety
2015
Shannon L. Ferrell, Esq., M.S.,
J.D.
Big data and drones on the
farm
Trademark
2019
Jane Wyngaard, Lindsay
Barbieri, Andrea Thomer, Josip
Adams, Don Sullivan,
Christopher Crosby, Cynthia
Parr, Jens Klump, Sudhir Raj
Shrestha and Tom Bell
Emergent challenges for
sciences UAS data
management: fairness through
community engagement and
best practices development
The lack of norms
or legacy UAS
data
2020
Roghieh Eskandari, Masoud
Mahdianpari, Fariba
Mohammadimanesh, Bahram
Salehi, Brian Brisco and Saeid
Homayouni
Meta-analysis of unmanned
aerial vehicle (UAV) imagery
for agro-environmental
monitoring using machine
learning and statistical models
Machine learning
and statistical
models
2019
Anam Tahir, Jari Böling,
Mohamma Hashem
Haghbayan, Hannu T.
Toivonen, Juha Plosila
Swarms of unmanned aerial
vehicles—a survey
Flying
mechanisms
2018
Gaurav Singhal, Babankumar
Shyam Bansod, Lini Mathew
Unmanned aerial vehicle
Remote sensing
classification, applications and
challenges: a review
2019
Miss. Alipta Anil Pawar, Dr.
Sanjay L. Nalbalwar, Dr.
Shankar B. Deosarkar, Dr.
Sachin Singh
Surveillance drone
Flight planning
BLDC (brushless
DC) motor
Ferrell et al. [10] investigated the common things among intellectual property
models to discard as a farm data protection tool for a trademark. Examples of trademarks is the basic design or the contoured bottle. One of the quickest approaches
is that the trademark fits in a model defining the farm information ownership for
branding purposes [10].
Wyngaard et al. [11] investigated the latest technology, the researchers are learning
how best to use UAVs. This gives a door of opportunity with the adopting new
application in a very minimal, and the net adoption of UAVs scientific data is very
small. However, these are very closing instantly, as many researchers are globally
creating all of the components for themselves in isolated manners, and accumulating
quickly data [11].
208
N. Vanitha et al.
Eskandari et al. [12] investigated machine learning in the general category under
machine learning methods. And the prediction models are being developed to establish a training data set which contains input data values. The semi-supervised methods
inspect the information contained in unlabeled data, which performance compared
to those on labeled data [12].
Anam Tahir et al., investigated UAV drones are divided into four main types
such as a fixed-wing hybrid. Fixed-wing, multi-rotor and single rotor. Fixed-wing is
mainly used for mapping and inspection of power lines flow over they can stay in air
for up to 16 h by using a gas engine [13].
Gaurav Singhal et al., investigated they’re of drones in agriculture for applying
amount for the given input to make products better agriculture has wide relate of
adopting the remote scoring technology using traditional ratel lite and aerial platform.
UAs has introduced a cheaper and low altitude alternative approach for providing
high- resolution images [14].
Miss. Alipta Anil Pawar et al., investigated battery because of uneven power
supply. To overcome this problem, he has used 188-li-fo battery voltage tester. This
tester repeatedly displays the battery’s voltage for our battery. It falls cell rct voltage
level battery for checking the round alarm [15].
4 Challenges of Big Data Analytics in UAV Surveillance
The following are the significant research challenges of the big data analytics in UAV
surveillance. A small drone may build even in novice using available parts over the
internet. Almost every small drone is in high safety risk like ground installations
parts of private property there are many occasional instances that are operated to loss
controls during UAV flight, so far, there is no serious accidents have been happened
using drones to supply illegal and banned items into prisons.
Government authorities have been testing to overcome those challenges with
appropriate regulations there are numerous rules and regulations for UAV operations
that are already using different technologies. This includes signal gamming as well
as attacking and capturing to trying the rogue UAV’s down. The drone industry is
getting advanced day by day large number of unmanned Aerial vehicles are being
used and sold all over the world. Private UAVs are crossing almost 10$ million dollars
in the first 10 years. Using such technologies are improving human living conditions.
4.1 Safety
In particular recreational drones and commercial unmanned aerial vehicles can count
the GPS technology for the searching process. This gives controllers to get accurate
and can read the vehicle location level remarkable distance. The GPS may fail to
Investigation on Challenges of Big Data Analytics …
209
awake controllers in nearby areas. UAVs shall interfere with the flight patterns and
potential safety threats [16].
4.2 Privacy
In besides to video surveillance, the systems can include sensors that they detect
magnetic fields and other chemical compositions, much more information. One of
the fundamental levels of the public can see the collection of data, almost all they are
being spied on. One of the good examples of this Drone technology is the Airship
and Transparency act provides briefly information on drone technology and how they
can leverage personal information about individuals.
4.3 Security
UAV has the potential to gather share the data as it possible that other sources can
attempt to hack the signals. The panic of sensitive data might end up in false hands,
may create frustrations. one of the best ways to ensure security of UAV is Radio
Frequency (RF) shielding [17].
5 Conclusion
Unmanned aerial vehicles are used to gather beyond a fixed range of sensors accurate
data over space and time. UAVs are quickly becoming a common part of a variety
of applications. In this chapter, we have briefly investigated the challenges of big
data analytics in UAV surveillance. However, there are growing new opportunities
for UAVs in the field of data processing and data worldwide. Particularly in northern
European countries, 59% of the share was published in 2018–2019. Hex copters are
being the most popular platform with a share of 30% in surveillance.
References
1. A.O. Agbeyangi, J.O. Odiete, A.B. Olorunlomerue, Review on UAVs Used for Aerial
Surveillance (Nigeria, 2016), pp. 4–7
2. J. Onohwakpor, The Use of Drone Technology as an Effective Tool in Providing Information
Services (Nigeria, 2020), pp. 8–12
3. A. Israr, G.E.M. Abro, M.S.A. Khan, M. Farhan, S.A.B.M. Zulkifli, Internet of Things (IoT)Enabled Unmanned Aerial Vehicles for the Inspection of Construction Sites: A Vision and
Future Directions (Pakistan, 2021), pp. 3–12
210
N. Vanitha et al.
4. N. Athanasis, M. Themistocleous, K. Kalabokidis, C. Chatzitheodorou, Big Data Analysis in
UAV Surveillance for Wildfire Prevention and Management (Cyprus, 2018), pp. 40–46
5. G. Sylvester, E-Agriculture in Action: Drones for Agriculture (Bangkok, 2018), pp. 57–76
6. M.A. Ma’sum, M.K. Arrofi, G. Jati, F. Anfin, M.N. Kurniawan, P. Mursanto, W. Jatmiko,
Simulation of Intelligent Unmanned Aerial Vehicle (UAV) for Military Surveillance (Indonasia,
2014), pp. 5–9
7. S.A. Shah, D.Z. Seker, S. Hameed, D. Draheim, The Rising Role of Big Data Analytics and
IoT in Disaster Management (Istanbul, Turkey, 2020), pp. 9–10
8. M.A. Khana, W. Ectorsa, T. Bellemansa, D. Janssensa, G. Wets, UAV-Based Traffic Analysis: A
Universal Guiding Framework Based on Literature Survey (Istanbul, Turkey, 2016), pp. 3–14
9. H. Hildmann, E. Kovacs, Using Unmanned Aerial Vehicles (UAVs) as Mobile Sensing Platforms
(MSPs) for Disaster Response, Civil Security and Public Safety (Heidelberg, Germany, 2019),
pp. 7–20
10. S.L. Ferrell, Big Data and Drones on The Farm (Ames, Iowa, 2015), pp. 5–12
11. J. Wyngaard, L. Barbieri, A. Thomer, J. Adam, D. Sullivan, C. Crosby, C. Parr, J. Klump, S.R.
Shrestha, T. Bell, Emergent Challenges for Science sUAS Data Management (USA, 2019),
pp. 8–13
12. R. Eskandari, M. Mahdianpari, F. Mohammadimanesh, B. Salehi, B. Brisco, S. Homayouni,
Meta-analysis of Unmanned Aerial Vehicle (UAV) Imagery for Agro-Environmental Monitoring
Using Machine Learning and Statistical Models (Canada, 2020), pp. 5–15
13. A. Tahir, J. Böling, M.H. Haghbayan, H.T. Toivonen, J. Plosila, Swarms of Unmanned Aerial
Vehicles—A Survey (Finland, 2018), pp. 3–7
14. G. Singhal, B.S. Bansod, Unmanned Aerial Vehicle Classification, Applications and Challenges
(Chandigarh, India, 2018), pp. 4–143
15. A.A. Pawar, S.L. Nalbalwar, S.B. Deosarkar, S. Singh, Surveillance Drone (Lonere, India,
2019), pp. 1–4
16. G. Woodhams, J. Borrie, Armed UAVs in Conflict Escalation and Inter-State Crisis (Geneve
Switzerland, 2018), pp. 8–14
17. M.H. Fleming, S.J. Brannen, A.G. Mosher, B. Altmire, A. Metrick, M. Boyle, R. Say,
Unmanned Systems in Homeland Security (Washington, DC, 2015), pp. 48–51
UAV-Based Photogrammetry and Seismic
Zonation Approach for Earthquakes
Hazard Analysis of Pakistan
Abdul Qahar Shahzad and Mona Lisa
Abstract Natural hazards have direct impact on the well-being of humanity in terms
of economy, infrastructure and sometimes leads to fatalities. Catastrophic event like
earthquake is considered as most devastating and quick phenomena. Therefore, there
is an urgent need for specifying earthquake-prone areas through Seismic Zonation.
Additionally, for spontaneous response to such emergency situation deployment of
FANETs based network are proposed. However, UAVs based system collect data
from post-seismic region in order to facilitate the victims. Also, real-time acquisition through Drones enhances rescue operation. Furthermore, proposed approach is
composed of Flying Ad hoc Networks and Seismic Zonation statistics which are
mainly utilized to improve surveillance in hazardous condition. Moreover, interpolation techniques are used to improve visual representation of seismic zone which
assist UAVs in life-saving tasks.
Keywords Flying ad hoc networks · Mobility models · Unmanned aerial
vehicles · Seismic zonation · Earthquake
1 Introduction
Geographically Pakistan is located in the region of active seismic zone especially
its Northern areas are highly impacted by regional warning, climatic crisis, prone
to earthquakes, high flooded area, severe landslides, droughts and high speed of
glaciers melting. Mainly in the past decades, Northwestern areas of Pakistan are
hugely affected by numerous tectonic events. Pakistan previous decade is full of
such disaster events where there were no proper reliefs or direct access were given to
victims. In case the government of Pakistan use the UAV during these catastrophically
events the impacts of earthquake might be mitigated through this latest approach of
A. Q. Shahzad (B) · M. Lisa
Department of Earth Sciences, Quaid-I-Azam University, Islamabad 45320, Pakistan
e-mail: shahzad@geo.qau.edu.pk
M. Lisa
e-mail: mlisa@qau.edu.pk
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_12
211
212
A. Q. Shahzad and M. Lisa
unmanned air vehicles like in October 2005 events, 2013 earthquake at Kashmir and
2015 respectively, besides this other various cities and town of Khyber Pakhtunkhwa
affected severely, millions of citizen were displaced, homeless and thousands of
people died even the infrastructure at some area was completely vanished. Ineffective
communication mechanism and cooperation amongst inhabitants and government
services, lacking community awareness, inadequate plan to maintain relation, poor
surveillance, complete absence of latest approach regarding technological (UAVs)
usage to build a pathway to tackle the situation. These were the prime and most
common challenges encountered by the existing relief teams and competent authority
of Pakistan to handle different disasters [1].
In recent years Pakistan disaster managements has launch new initiatives regarding
these seismic events to prevent as much as possible by providing an effective communication system along with continuous monitoring of natural force affected areas
especially the earthquake. National disaster management authority (NDMA) with
the collaboration of military aeronautical wings placed the drone in Northern areas
for continuous surveillance in seismic prone areas. In case of any emergency these
UAV are mainly responsible to capture the real time image of area such as roads,
bridge condition, buildings, and other infrastructure. UAV-based Photogrammetry
and Geocomputing provide a direct access to affected area in terms of hazard analysis and feasibility assurance whether the victims could be assisting through ground
or air. Another significant aspect of UAV flights is maintaining the main network and
communication channel [2].
2 UAVs Impacts in Hazard Analysis and Rescue-Based
Mission
Although the post-earthquake spatio-temporal frequency has restricted the use of
UAVs in research work, it provide valuable post-earthquake survey tool to collect
data, for such areas which are not safe to access or monitor especially damaged
building and infrastructure [3, 4]. As the UAVs are complement to higher scale
manned-aircraft photography and photogrammetry, so they are efficient to offer quick
solutions with virtually no direct need of infrastructure such as airport, runway, and
airspace control facility. Even UAVs are considered as idealistic approach to detect
mini-scale alterations, ground rupture, rock fracture, fault movement and cracks
in natural and human-made structures. Indeed, UAVs usually provide numerous
data with enhanced resolution and most significantly a low-cost solution to affected
area although UAVs can’t cover large areas. During the recent seismic activity in
Kumamoto earthquake in south Japan (Kyushu Island) UAVs pattern was successfully conducted, while this post-earthquake the administration use UAVs in smallscale survey to detect surface rupture of the active fault displacement. Furthermore the
competent authorities of Japan disaster management use the UAVs to detect the structural changes in historical building which was difficult to monitor from the traditional
UAV-Based Photogrammetry and Seismic Zonation Approach …
213
aircraft, like roof conditions, collapsed walls and construction sites surrounded by
large trees which prevent the NADIR view. UAVs images captured from the affected
areas can easily convert into 3D models for accurate post-earthquake impacts. Similar
technique was applied in L’Aquila earthquake in Italy by using UAVs collective with
the photogrammetric approach of SFM [5]. Even the UAVs are capable to collect
data which are beyond human visual observations. For instance, the Japanese disaster
authorities have successfully used the UAVs to inspect the Fukushima power plant
from above and around to analyze the active radiation levels after the incident. Beside
continuous surveillance UAVs were responsible for the transportation of emergency
goods, such as first aid kit, and other medical aids [6], plus food supply with other
vital items, especially when there is no direct access to area due to collapsed roads
and communication services [7, 8].
3 Seismicity of the Area
Pakistan located in the active seismicity and primarily its Northern region are prone
to earthquake even its adjoining countries such as Afghanistan, Iran, China and India
experiences high intensity/magnitude of earthquake Consequently these tectonic
activity leads to huge losses of life and infrastructure respectively. Additionally the
active fold and thrust belt namely north-west Himalayas and Sulaiman ranges also
plays a significant role in the seismicity of the region mainly focus on the Makran
coastline events of year (1945), which having Mw (8.3) such seismic events result in
numerous small islands alongside with the Makran coastal belt [8]. Along the neighboring country India also experienced analogous earthquake of Kangra (1905) having
magnitude Mw (7.8) activating the Main Boundary Thrust (MBT). This MBT is the
leading regional fault in surrounding area. Furthermore in the recent years, (Pattan
1974), (Rawalpindi 1977) (Bunji 2002), (Batgram 2004) and (Kashmir 2005) in
which earthquake severely affect the study area [9, 10].
4 Regional Tectonic Setup
The Himalayan Orogeny is the youngest mountain terrains around the globe which is
the product of Indian and Eurasian plate’s collision. The site of collision is considered
to be a broader zone and marked by the Himalayan-Kurrakuram-Hindukush ranges
in northwestern Pakistan. The Himalayan ranges along with adjacent mountain chain
trend E-W and switching more N-S trend in the west [11].
Starting about 55 Ma ago, collision of the Indo-Pak Sub Continent with Eurasia
Plate produced the Himalayan Orogenic Belt which includes metamorphic and
igneous rocks of the South Asian plate emplaced over the rock of Kohistan arc along
main Karakuram Thrust (MKT) at time of late cretaceous [12]. Kohistan Island arc
(KIA) terrain is composed of metamorphic and basic to ultra-basic rocks which are
214
A. Q. Shahzad and M. Lisa
thrusted over north side of Indian Plate along the MMT (Main Mantle Thrust) [13].
The foreland consisting of telescoped igneous, sedimentary and metamorphic rocks
of the Indian foreland basin is marked to the South by (MBT). The MBT extends
westward from the front of main Himalayan range around Hazara Kashmir syntaxes
and thrust the hill ranges over the Kohat-Potwar foreland basin. The main boundary
thrust system contain intensely deformed Pre Cambrian to Cenozoic sedimentary
rock which is younger to the south. The Plio-Pliestocene sedimentary rocks of Kohat
foreland basin is also deformed bounded by SRT in Potwar and southern side of
Trans Indus Salt Ranges while it is undeformed in Bannu basin. The un-deformed
foreland area or indo-gigantic plain lies to the south of the SRT and Trans-Indus
ranges and is the current day depo-center for the Himalayan shed.
The northwest Himalayas in Pakistan can be further classified into three main
tectonomorphic terrains which are divided by major thrust fault systems.
4.1 Main Karakoram Thrust
It is the northernmost member of thrust system of northern Pakistan. The Karakuram
Thrust separates the Kohistan Island Arc from the metamorphic and igneous
complexes of south Asian Plate to the North. Kohistan Island Arc terrain is composed
of metamorphic and basic igneous rock which is thrusted over the northern margin
of Indian Plate along the MMT. Kohistan Island Arc sequence was sutured to Asian
Plate along MKT. whereas, it is believed that ocean between the Karakuram and the
Kohistan Arc is closed in the late Cretaceous [14].
4.2 Main Mantle Thrust
MMT is the product of the Subduction of Indian Plate beneath Kohistan Island arc.
This collision between Indian plate and the Kohistan island arc took place in the
Eocene [12].
4.3 Main Boundary Thrust
The Main Boundary Thrust (MBT) extends west ward from the front of Himalayan
Range and Hazara Kashmir Syntaxes and thrust the hill range over Potwar Kohat
foreland basin. The MBT system composed of intensely deformed Precambrian
to tertiary sedimentary rocks which continuously becomes younger to southward.
Active faulting occurred along the Main Boundary Thrust to north along East-NorthEast-trending high angle fault and adjacent pressure ridge in southern Peshawar basin
[15].
UAV-Based Photogrammetry and Seismic Zonation Approach …
215
5 Neighbor Embedding for Seismic Zonation
The following method can be used to resolve the issue of single image resolution.
With the input of the χt as an image of low-resolution, targeted enhanced resolution image (γt ) using with instruction of one or more image set of low-resolution
(χs ) and equal to image of high-resolutions (γs ) can be determined [16, 17]. Each
image whether it is categorized as low or enhanced resolution can be signified
as combination of small overlying image areas. χt and γt has identical patches
number, and each χ , γ also do same. All these parameters can be denoted as
q Nt
p Ns p Ns s qs Nt
.
X s p=1 , γs p=1 , X t q=1 and γt q=1
Apparently, Ns and Nt are depending on degree and size of overlapping concerning
neighboring patches. Preferably, every patches created for image of higher resolution
(γt ) should not only be connected properly to the corresponding patch in the low resolution image X t , it must tolerate some of the inter-patch contacts with neighboring
areas in Yt . Previous estimates the precision while the latter calculated the local
compatibility and smoothness. Possibly, to gratify these necessities, the following
properties might be our way: (a) γt has numerous patch alterations cultured from
the instruction set. (b) Local associations between areas in X t should be preserved
in γt . (c) Nearby patches in γt are inhibited through overlapping to enforce local
compatibility and smoothness [16].
5.1 Manifold Learning
Supposed system is specifically based on these small patches in the low and high
resolution images form manifolds with comparable local-geometry in two distinctivespaces. Since, the ensuing depiction is constant and with no resolution if the embedding is isometric, this statement might be suitable. Each patch denotes characteristic vector, with corresponding two characteristic spaces. For handiness, we
p
p
q
q
use X s , ys , X t and yt to indicate the characteristic vectors along with equivalent
image patches, X s , ys , X t and yt to indicate the groups of characteristic vectors with
equivalent images [18].
In current time, limited new manifold learning (or decreasing of non-linear dimensionality) means have been strategized to automatically determine low-dimensional
non-linear manifolds in high dimensional data set in them onto low dimensional
embedding spaces, using tractable linear algebraic techniques that are not prone to
local minima. These cover isometric characteristic mapping (Isomap), locally linear
embedding (LLE), and Laplacian eigenmap. Our incredible-resolution techniques to
be defined below have been stirred by these manifold learning techniques, chiefly
LLE [17].
216
A. Q. Shahzad and M. Lisa
5.2 Seismicity of Pakistan
Pakistan northern areas are highly prone to earthquakes. Figure 1 represents entire
events recorded from various observatories. Figure 2 illustrates earthquakes records
based on depth from 0 to 50 km, where Fig. 3 shows events up to 100 km depth.
Additionally, Fig. 4 exhibits seismic events from 101 to 250 km depth.
5.2.1
Interpolation Result Discussion Based on Neighbor Technique
The cluster data of the earthquake’s events are interpolated through Near Embedding
techniques and magnitude maps were generated. The magnitude maps for all these
events shows that most of the earthquake events are of moderate magnitude in the
northern part of Pakistan. This is probably due to the reasons that the northern part of
Pakistan demarcates and lies in the vicinity of the tectonic boundary between Indian
and Eurasian plate. These figures also show that the Quetta syntaxes and its active
faults (e.g. Takhatu Fault, Ziarat Fault, Chilton fault, Chaman fault, Sariab faults,
Bibai fault etc.) also have moderate magnitude. Some events of high magnitude are
also observed in both the northern and southern parts of Pakistan. The major faults
in the northern region of Pakistan are Main Karakorum Thrust (MKT), Main Mantle
Fig. 1 Map depicting the frequency on regional scale all over Pakistan that varies from very high
to lowest values
UAV-Based Photogrammetry and Seismic Zonation Approach …
217
Fig. 2 Map showing frequency on regional scale from 0 to 50 km, where again Northern areas
have highest values and covers more area than other
Fig. 3 Map showing frequency on regional scale from 50 to 100 km, where again Northern areas
have highest values and covers more area than other
218
A. Q. Shahzad and M. Lisa
Fig. 4 Map showing frequency on regional scale from 100 to 250 km, where again Northern areas
have highest values and covers more area than other
Thrust (MMT) etc. All these events in the northern part are associated with these
faults.
6 Bilinear Interpolation for Seismic Zonation
Recent studies on image interpolation have greatly enhanced our knowledge for
various applications. Its main purpose is the perceptual worth of images. That is, the
interpolated images should be artifact-free and visually understandable. Many recent
image interpolation techniques have been built for the aforementioned purpose. These
methods, generally termed as content-adaptive, edge-directed, and/or level-set based.
Current work uses the visually-oriented-methods because it is most widely applicable
and has fast-growing demands for processed images/video of the photo-cartography,
multimedia, TV, printing, and advertising industries. These visually oriented interpolation techniques falls in the category of an orientation-adaptive-interpolation
method. Bilinear and Bicubic methods are the most widely and traditionally used
amongst them. The pitfall of both the bilinear and bicubic methods includes several
types of meandering artifact, and the visual degradations. The Meandering artifacts
are called the jaggies and are commonly appear as a stair-casing of image edges in
UAV-Based Photogrammetry and Seismic Zonation Approach …
219
areas of fine textures/ground. Since edges are characterized by artifacts, amongst
them meandering is most effective and influencing [19].
A considerable array is there in the methodologies for reducing meandering
between these recent visual based oriented interpolations techniques. The isophoteoriented methods are the projected one amongst them. Isophotes are defined as the
equi-intensity-contours (also called as level-set contours). All methods treat interpolation as a variational problem. A restriction to the isophotes is given in these
interpolated images in order to have minimum curvature and is introduced within the
interpolation algorithms. The curvatures of the isophotes are “smoothed,” through
solution of partial differential equations (PDEs) and meandering artifacts are thus
reduced. The interpolated final image should have minimum error in image gradientangle at the points of original pixels. In response of this mentioned restriction along
with some assumptions, the obtained PDEs can be simplified in a one-pass-discrete
shape. The obtained results show that the isophote-oriented techniques are more
precise in the reducing meandering artifacts in the final processed images. However,
some of its further disadvantages are also there which have to be overcome e.g. the
subsequent [20, 21].
Some of these methods introduce several undesirable artifacts/errors to processed
interpolated maps, for example, the “cartooning” effects related to the edge-directed
methodologies. Several of these techniques have the restriction to its expansion ratio
or, where an integer lies. Our investigation delineates some of these fresh visually
oriented methods have the aforesaid drawbacks.
Moreover, besides the aforementioned visually oriented methods, some other
interpolation techniques i.e. the shape-based, B-spline-based, and registration-based
method, are also entered to the recent research awareness. Their algorithms have
achieved hopeful improvements also over the traditional cubic techniques and linear
one. Here, not all the recently explored methods are proposed but only is focused on
applying the visually oriented methods only [21].
Here onward a new isophote-based, orientation-adaptive-method is recommended. The selected technique is directional one and the interpolation kernel
operates on the local orientation of isophotes. Such approaches towards the results
obtained in the interpolated images have much smoother isophotes with their orientations, thus considerably reducing meandering artifacts. Secondly, it describes the
onward orientation-adaptive-interpolation way. Thirdly, we discuss the interpolation
of the ridges in an image. Then the isophotes are built by using the bicubic, bilinear,
and other proposed methods. After that, the resulted outcomes show the quantitative
evaluations [18].
6.1 Directional Bilinear Interpolation Approach
In the conventional bilinear interpolation approach, the weightage of a new (child)
pixel is incorporated from its four nearest original parent pixels. The mathematical
weightage of a child pixel is given by
220
A. Q. Shahzad and M. Lisa
F(x, y) = A + (B − A) ∗ x + (D − A) ∗ y + (A + C − B − D) ∗ xy.
(1)
Here, the indices of the pixels (i.e. A, B) refer to the intensity standards,
if their authentic significance is palpable from the framework. After that, the
content-independent square-interpolation kernel engaged and linked by traditional
bilinear technique which causes the isophotes fan interpolated image squiggly, thus
results meandering artifacts. Our preceding work gives attention to this customary
bicubic-interpolation which has also similar zigzagged isophotes [18, 22].
The adapted interpolation method is able to recover the traditional bilinear-method
by incorporating the orientation-adaptive-parallelogram-based interpolation-kernel.
The strength value of the child pixel is then extrapolated by using the parent
pixel slaying on its vertices which is considered to be the vertices of a parallelogram and the child pixel encloses. The course/texture and its shape diverge depends
on its local route of the isophote via child pixel. The method which decides the
parallelogram-kernel is introduced in the feature through subsequently segment. In
every parallelogram, the worth of the child pixel is estimated by using a form similar
to (1)
F(x, y) = A + αx + βy + γx y .
(2)
where, x and y are the distances which can be calculated down from the orientation
of parallelogram?
x =
x
and y = y − x ∗ tan θ.
θ cos
(3)
In Eq. (2), coefficients α, β, and γ are defined as
α = (B − A) cos θ, β = D − A, γ = (A + C − B − D) cos θ.
(4)
Note that, in special case, the interpolation by a parallelogram becomes equal to
the interpolation obtained from a square kernel as in Eq. (1) [23].
6.2 Results and Interpration from Pakistan Bilinear-Based
Interpolation
The cluster data of the earthquake’s events are interpolated through Bilinear Interpolation techniques and magnitude maps were generated. The magnitude maps for
all these events shows that most of the earthquake events are of moderate magnitude in the northern part of Pakistan. This is probably due to the reasons that the
northern part of Pakistan demarcates and lies in the vicinity of the tectonic boundary
between Indian and Eurasian plate. These figures also show that the Quetta syntaxes
and its active faults (e.g. Takhatu Fault, Ziarat Fault, Chilton fault, Chaman fault,
UAV-Based Photogrammetry and Seismic Zonation Approach …
221
Fig. 5 Map shows regional zonation of all earthquake catalogue events
Sariab faults, Bibai fault etc.) also have moderate magnitude. Some events of high
magnitude are also observed in both the northern and southern parts of Pakistan
[24, 25].
Additionally, Fig. 5 represents all earthquakes events recorded. The catalogue
contains approximately 32,000 events. Figure 6 shows shallow earthquakes whereas
Fig. 7 represents moderate seismic events. Moreover, deepest earthquake zonation
is illustrated in Fig. 8.
7 Conclusion
Although in the last few years mainly the UAV technological approach has only utilize
for the commercial purposes, its efficiency, quick action, safe access to dangerous
sites, low-cost solution and convenient accessibility is highly expected to further
boost its speedy growth, aims to provide support and emergency services in affected
region of earthquake consequently, the chance to implement UAVs in disaster areas
where direct accessibility through road is difficult plus not being restricted to use in
numerous scientific experiments. In recent years UAVs present huge contribution in
hazard analysis, monitoring, risk management, and emergency services, as well its
application is also proved successfully in other geo-hazards: earthquakes, flooding’s,
volcanic eruptions, rock fall related activities, and landslides. UAVs usages in such
222
A. Q. Shahzad and M. Lisa
Fig. 6 Map shows regional zonation from ground surface to 50 km depth
Fig. 7 Map shows regional zonation from 50 to 100 km depth
UAV-Based Photogrammetry and Seismic Zonation Approach …
223
Fig. 8 Map shows regional zonation from 100 to 250 km depth
disaster areas have emphasized the direct link survey-scale (spatio-tempory, mini and
micro scale, time and space) and the scale of natural phenomena. Seismic zonation
provides us the future probabilistic approach whether the area is prone to seismic
activity or the area is relative safe than surrounding area. Such mathematical approach
and seismicity pattern integrates the region according their previous events which
further assist the UAVs to monitor the urban areas in case of any seismic activity.
Consequently the UAVs have been used in post-earthquakes for risk analysis, continuous surveillance and emergency services, because seismic activities earthquakes are
greatly unpredictable in time. Future work is combined photogrammetry with electromagnetic precursor of earthquake using the remote sensing technology, in order to
capture and measure the rapid changes in anomalies using as an earthquake precursor.
Although several reconstruction based on one or numerous tools using algorithms are
present theoretically, it implementation in real time (3D or 4D measure) haven’t been
performed yet and these finding could be very significant to understand upcoming
seismic events and other natural phenomena related to geosciences such as anomalies.
References
1. H.U. Salam et al., Drone based resilient network architecture for survivals in earthquake zones
in Pakistan. Sindh Univ. Res. J. Sci. Ser. 50(01), 175–182 (2018)
2. V. Baiocchi, D. Dominici, M. Mormile, UAV application in post-seismic environment. Int.
224
A. Q. Shahzad and M. Lisa
Arch. Photogram. Rem. Sens. Spat. Inf. Sci. 1, W2 (2013)
3. R. Wodak, M. Meyer (eds.) Methods of Critical Discourse Studies (Sage, 2015)
4. K. Kirihata et al., Fabrication of copper/single-walled carbon nanotube composite plating films
by electrodeposition, in ECS Meeting Abstracts, vol. 47 (IOP Publishing, 2015)
5. J. Liu, J. Coco et al., Particulate air pollution from wildfires in the Western US under climate
change. Clim. Change 138(3), 655–666 (2016)
6. C.A. Thiels, et al. Use of unmanned aerial vehicles for medical product transport. Air Med. J.
34(2), 104–108 (2015)
7. A. Nedjati, B. Vizvari, G. Izbirak, Post-earthquake response by small UAV helicopters. Nat.
Hazards 80(3), 1669–1688 (2016)
8. A. Nedjati et al., Complete coverage path planning for a multi-UAV response system in postearthquake assessment. Robotics 5(4), 26 (2016)
9. G.-J. Jiang et al., An improved multi-criteria emergency decision-making method in environmental disasters. Soft Comput., 1–29 (2021)
10. A. H. Kazmi, M. Q. Jan, Geology and Tectonics of Pakistan (Graphic Publishers, Karachi,
1997), p. 554
11. M. Zaré, S. K. Paridari, Balakot, Muzaffarabad Earthquake of 8 October 2005, Mw 7.6; field
observations on geological aspects, in The 14th World Conference on Earthquake Engineering
(Beijing, China, 2008)
12. B. Baldwin, C.O. Butler, Compaction curves. AAPG Bull. 69(4), 622–626 (1985)
13. M. Zaré, S. Karimi-Paridari, An investigation on Balakot, Muzaffarabad (Pakistan) earthquake,
8 Oct. 2005, Mw 7.6; geological aspects and intensity distribution. J. Seismolog. 13(3), 327–337
(2009)
14. H. Ghani et al. Structural evolution of southern Kohat fold and thrust belt, a case study from
Karak area, Pakistan. J. Himalayan Earth Sci. 48(2), 1 (2015)
15. B.N. Cox, R.A. Tahir-Kheli, R.J. Elliott, Thin films of itinerant-electron ferromagnets on a
nonmagnetic metallic substrate. Phys. Rev. B 20(7), 2864 (1979)
16. P.J. Treloar et al., K-Ar and Ar-Ar geochronology of the Himalayan collision in NW Pakistan:
constraints on the timing of suturing, deformation, metamorphism and uplift. Tectonics 8(4),
881–909 (1989)
17. H. Chang, D.-Y. Yeung, Y. Xiong, Super-resolution through neighbor embedding, in Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern
Recognition, 2004. CVPR 2004, vol. 1 (IEEE, 2004)
18. K.T. Gribbon, D.G. Bailey, A novel approach to real-time bilinear interpolation, in Proceedings. DELTA 2004. Second IEEE International Workshop on Electronic Design, Test and
Applications (IEEE, 2004)
19. X. Gao et al., Image super-resolution with sparse neighbor embedding. IEEE Trans. Image
Process. 21(7), 3194–3205 (2012)
20. S. Fadnavis, Image interpolation techniques in digital image processing: an overview. Int. J.
Eng. Res. Appl. 4(10), 70–73 (2014)
21. N. Jiang, L. Wang, Quantum image scaling using nearest neighbor interpolation. Quantum Inf.
Process. 14(5), 1559–1571 (2015)
22. P.R. Smith, Bilinear interpolation of digital images. Ultramicroscopy 6(2), 201–204 (1981)
23. C. Accadia et al., Sensitivity of precipitation forecast skill scores to bilinear interpolation
and a simple nearest-neighbor average method on high-resolution verification grids. Weather
Forecast. 18(5), 918–932 (2003)
24. C. Gomez, H. Purdie, UAV-based photogrammetry and geocomputing for hazards and disaster
risk monitoring—a review. Geoenviron. Disast. 3(1), 1–11 (2016)
25. M. Lisa, A.A. Khwaja, M.Q. Jan, Seismic hazard assessment of the NW Himalayan fold-andthrust belt, Pakistan, using probabilistic approach. J. Earthq. Eng. 11(2), 257–301 (2007)
Optimizing UAV Path for Disaster
Management in Smart Cities Using
Metaheuristic Algorithms
Zakria Qadir, Muhammad Hamza Zafar, Syed Kumayl Raza Moosavi,
Khoa N. Le, and Vivian W. Y. Tam
Abstract In this research, different state-of-the-art metaheuristic algorithms are
examined in order to incorporate a collision-free path planning technique for
Unmanned Air Vehicles (UAV) in crisis situations. The effective path planning is
utilized to identify the major cause of disasters such as bushfires, which are wreaking
havoc on the global forest environment. This new approach is a first step toward a predisaster evaluation and options for saving survivors in a shorter amount of time. For
UAV path optimization, a novel meta-heuristic algorithm i.e. Smart Flower Optimization Algorithm (SFOA) is presented. Comparison is made between different metaheuristic algorithms such as Particle Swarm Optimization (PSO), Flower Pollination
Algorithm (FPA) and Grasshopper Optimization (GHO). Four different scenarios
are offered to demonstrate the resilience of our proposed model that are dynamic
environment (DE), general environment (GE), condensed environment (CE) and
maze environment (ME). The barriers in each scenario are put in such a way that the
overall path complexity for a UAV to reach the destination is increased. In SFOA, two
growth methods are controlled on the movement of immature flower and mathematical modelling is used to update particles position. The SFOA method beats other
algorithms based on the parameters chosen, saving up to 24.5% in transportation
costs and 13.3% in computational time. As a result, SFOA can be used to optimize
UAV path planning in any of the foregoing environment.
Z. Qadir (B) · K. N. Le · V. W. Y. Tam
School of Engineering, Design and Built Environment, Western Sydney University, Locked Bag
1797, Penrith, NSW 2751, Australia
e-mail: z.qadir@westernsydney.edu.au
K. N. Le
e-mail: k.le@westernsydney.edu.au
V. W. Y. Tam
e-mail: V.Tam@westernsydney.edu.au
M. H. Zafar
Capital University of Science and Technology, Islamabad 44000, ISB, Pakistan
S. K. R. Moosavi
National University of Sciences and Technology, Islamabad 44000, ISB, Pakistan
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_13
225
226
Z. Qadir et al.
Keywords UAV path planning · Disaster management · Smart cities ·
Metaheuristic algorithm · SFOA
1 Introduction
In this period of technical growth, UAVs (unmanned aerial vehicles) are becoming
more prevalent and attracting attention particularly in the disaster management
sectors, traffic inspection and cargo [1]. Different path planning algorithms are
assisting numerous applications in the realm of disaster management for smart cities
[2–4] by analyzing and proposing a resilient route for these UAVs [5, 6]. However,
as illustrated in Fig. 1 [7], determining an optimum path in unknown situations
when there are a large number of obstacles in the path is a major difficulty. Several
methods for path planning have been developed by many researchers, all of which
provide a collision-free environment [8–10]. Despite technical advancements, it is
difficult to manage the solution’s efficiency and quality, particularly when there
are several impediments and uncertainties in the environment [11]. Several problems have been solved using metaheuristic algorithms for this purpose [12–14].
For modern optimization problems, however, there are no such metaheuristic algorithms that can be deemed 100% flawless [15]. The process of determining the best
combination of parameters and variables to fulfil the goals is known as optimization.
Efficiency, profit, and cost are examples of objective functions. It entails minimizing
and maximizing the goal function, as well as determining the best solution [16].
For four reasons, metaheuristic algorithms are becoming increasingly popular in
optimization [17–20]. To begin with, metaheuristic algorithms are straightforward.
Their simplicity stems from the fact that they are based on basic notions. They’re
usually linked to natural physical phenomena like animal behavior and evolution,
Fig. 1 UAV path planning approach for disaster management
Optimizing UAV Path for Disaster Management in Smart Cities …
227
for example. Scientists should learn meta-heuristic algorithms so that they can apply
them to their challenges because of their simplicity. Because they are so simple, scientists can employ natural concepts to suggest new heuristic algorithms and improve
existing metaheuristics.
The second reason is that they are adaptable. Metaheuristic algorithms can be
applied to a variety of settings and problems without requiring any changes to the
methods. Meta-heuristic algorithms are easily adaptable to a variety of contexts and
difficulties because they treat problems as black boxes. The most important thing a
computer scientist has to know about these algorithms is how to express a problem
to them.
Third, most meta-heuristic algorithms have mechanisms that do not require derivation. Meta heuristics use stochastic optimization to solve various problems. It indicates that optimization begins with a random solution, and we don’t need to calculate
the derivative to get the best answer. As a result, meta-heuristic algorithms are best
suited to real-world issues. The finest thing about their adaptability is that they may
be used in a variety of fields.
Fourth, metaheuristic algorithms outperform traditional optimization strategies in
terms of avoiding local optima. All of this is due to the stochastic character of these
algorithms, which allows them to search the entire search space extensively and avoid
local solutions stagnating. Because the search space for analysing the pre-bushfire
catastrophe assessment is complex and uncertain, metaheuristic algorithms are an
excellent choice for difficult tasks like the one we propose in this study.
There are several metaheuristic algorithms accessible, but in this work, we are
primarily interested in and concerned about SFOA, as well as some other algorithms
such as PSO, GHO, and FPA [21–25]. In essence, a SFOA is an algorithm that is
influenced by individual cooperation in order to reach an optimization goal. It has
been put to the test on a wide number of mathematical optimization problems, and the
findings have been confirmed using the most up-to-date available techniques. These
findings demonstrate the algorithm’s high level of engagement. The phenomena
of how group members are organised into subgroups and fulfil distinct jobs with
collaboration influences the Dynamic Group-based algorithm. This SFOA algorithm
will divide persons into subgroups to accomplish their stated jobs in the case of
unmanned vehicles.
PSO refers to the probabilistic optimization algorithm. The distinctive behavior
of creatures that live in groups, such as birds and fish, influences our populationbased approach. Then there’s FPA, which is primarily concerned with grey wolf
behavior, such as leadership and hierarchy. It is frequently utilized in applications
such as future selection, robotics, pathfinding, clustering, and so on. The humpback
whales’ bubble net foraging has an impact on the overall performance. They have
large computing costs and are utilized for work scheduling, laser sensor accuracy
control, route optimization, and other applications. GHO is based on barnacles, which
have been around since the Jurassic period. It is based on the evolution process, in
which a Global-optima is created by producing new offspring with features and
characteristics acquired from the parents. This algorithm’s solution set is enhanced
until the desired terminating condition is achieved.
228
Z. Qadir et al.
In this article, a unique Smart Flower Optimization (SFOA) algorithm is employed
to optimize UAV path planning. The SFOA’s most important features are:
• Limited number of random variables and tuning parameters.
• Effective exploration and exploitation phase enhance the global optimum solution
finding capability.
2 Related Study
Meta-heuristic optimization algorithms have gotten a lot of interest in the last decade
because of their capacity to solve real-time engineering challenges through effective
exploration and exploitation. Meta-heuristic optimization algorithms have a wide
range of applications, including neural network training, path planning and renewable
energy power extraction and so on. The ability of meta-heuristic algorithms to find
global minima is related to the inherent unpredictability of particle updating. Single
agent-based optimization algorithms search for global optima with a single particle,
which takes a long time to search the whole search space for the single particle.
Simulated annealing, genetic algorithms, and other single agent-based algorithms
are examples.
The multi-agent-based optimization method, which has more than one particle,
is another type of optimization algorithm that depends upon the application. particle
swarm optimization with gravitational search, Particle swarm optimization (PSO),
cuckoo search (CSA), grey wolf optimizer (GWO), whale optimization algorithm
(WOA), artificial bee colony (ABC), dragonfly optimization (DFO) and grasshopper
optimization algorithm (GHO) are some of the multi-agent-based algorithms available (PSOGS) [26–30]. Random numbers are embedded in the velocity vector in
particle swarm optimization, which generates random oscillations and does not
successfully settle at global minima. The grey wolf optimizer’s decreasing value of
parameter “a” promotes less explorative activity over iterations, resulting in particle
settling prematurely. Another attempt to address the oscillations problem in PSO is
to include a gravity effect (PSOGS), although this additional impact restricts particle
movement in search space. In DFO, dynamic grouping of males and females leads
gradual convergence to global optima, making the optimization issue less practical.
GHO features a parameter called “c” that balances the exploration and exploitation tendencies. This parameter creates oscillations and takes a long time to reach a
convergent state. The probability function in ABC is used to maximize the exploitation of random selection of bad-solutions. In the exploitation phase, the Levy flight
random walk for updating particles in CSA creates oscillations, which can easily
de-track from the global optima, and does not settle properly.
As we all know, path planning is a crucial responsibility for UAV optimization nowadays, and many algorithms have been presented as the best for achieving
maximum efficiency in this regard, including, PSO, FPA and GHO [31–34]. However,
when all of the algorithms’ findings are pooled, SFOA emerges as the best and most
cost-effective metaheuristic method to date. As an example, an algorithm like PSO is
Optimizing UAV Path for Disaster Management in Smart Cities …
229
incredibly crucial for path planning in complicated environments, but it is also very
time demanding. So as the other algorithms like, FPA and GHO.
Atif et al. [35] proposed a method for linear scaling of the search area in relation
to noise intensity. The cost of UAV energy was decreased by 70% using this strategy.
The author also suggested a mathematical approach that may be combined with
PSO to produce a feasible UAV path planning solution. Wen et al. [36] proposed an
online path planning system for low-altitude environments with dense obstacles and
difficult path finding. As a result, a novel method was investigated to determine the
safest path. Wahab et al. [37] developed a hybrid PSOFS algorithm for mobile robot
path planning in a complicated indoor environment. Gul et al. [38] employing FMH-,
developed an integrated approach for multi robot exploration by merging the CME
and GHO, and that study aided in exploring unexplored regions in the environment.
Savkin and Huang [39] investigated the problem of path planning for aerial drones
monitoring disasters and suggested an efficient path planning algorithm. Cho et al.
[40] proposed a wireless sensor that could be recharged to extend its life. To achieve
more flexibility, a self-propelled vehicle was paired with a charger. Rossello et al. [41]
also keeps an eye on a distributed dynamical environment for an information-based
mission planner with a UAV.
2.1 Abbreviations
The list of symbols and their abbreviations are shown in Table 1.
Table 1 List of defined
symbols
Symbol
Abbreviation
x(i)
Current position of UAV
M
Number of immature sunflowers
Dim
Number of variables (dimension)
L Itr
best,SF
Current best length of the sunflower’s stem
Itr
Current iteration
d
Damping
φ
Phase angle
f(x)
Cost function
Ps
Starting point
Pe
End point
P1 , P2 , P3
Control points
230
Z. Qadir et al.
3 Mathematical Modeling and Metaheuristic Algorithm
For UAV path planning, our suggested model integrates various metaheuristic techniques. The SFOA algorithm and the two stages utilized to solve complex path
planning scenarios are discussed in length in this section [42, 43].
Smart Flower Optimization Algorithm (SFOA) is a population-based optimization
algorithm. The immature sunflower’s development mechanisms can be used to update
search agents for the SFOA. Each immature sunflower in a Dim-dimensional search
space is considered to have a stem length in this algorithm. As SFOA is a populationbased algorithm, the set of immature sunflowers can be represented as follows in a
matrix [3, 44]:
⎡
s f 1,1 s f 1,2
⎢ s f 2,1 s f 2,2
⎢
SF = ⎢ .
..
⎣ ..
.
s f M,1 s f M,2
⎤
· · · s f 1,Dim
· · · s f 2,Dim ⎥
⎥
⎥.
..
..
⎦
.
.
· · · · · · s f M,Dim
···
···
..
.
(1)
where, M stands for the number of immature sunflowers and Dim stands for the
number of variables (dimension). The stem length of each immature sunflower represents a solution to the optimization problem in the proposed SFOA. Each sunflower
has a fitness value that corresponds to the value of the optimization problem’s objective function, which shows the length of its stem. The longer the stem of the sunflower,
the higher the fitness score. The internal systems that allow the sunflowers to forecast
and prepare for finishing their growth throughout the new day in the decision space
are used to create new stem length (solutions). These internal systems are controlled
by the solar clock during the day and the biological clock at night. The attributes of the
SFOA are listed in Table 1. The mathematical model used to replicate the immature
sunflower’s growth mechanisms is offered in two modes: bright and cloudy or rainy.
Toggling between sunny and cloudy modes was done with the parameter “Sun.” It is
set to ‘1’ when the weather is sunny and ‘0’ when the weather is cloudy or rainy. For
this study, the first mode has been chosen as the default mode. The proposed SFOA’s
first mode is represented by the following equation:
L Itr+1
new,SF =
Itr
Itr
L Itr
old,SF + d × sin(ω) × Aux × L best SF − L old,SF , hour sday ≤ 24
Itr
Itr
It
L old,SF + d × sin(ω) × L best,SF − L old,SF , other wise
(2)
where, L Itr
best,SF indicates the SFth element of the current best length of the sunflower’s
stem at Itrth iterations. In nature, the heliotropic movements of an immature sunflower
gradually reduce until the sunflower achieves maturity, at which point it completely
ceases to move. The parameter ‘d,’ which stands for damping, was employed in
SFOA to approximate the end of sunflower stem elongation. During iterations, the
parameter ‘d’ is adaptively lowered using the following equation:
Optimizing UAV Path for Disaster Management in Smart Cities …
d = dampingmax − Itr ×
damping max − damping min
Itrmax
231
(3)
where the maximum and minimum values of damping parameter are dampingmax
and dampingmin . Current iteration is denoted by Itr, and the maximum number of
iterations are denoted by Itrmax . The heliotropic motions of immature sunflowers
in response to the ‘Auxin’ during a 24 h day/night cycle are described by the sine
function.
The parameter ‘ω’ specifies the angle at which the stem of the sunflower should
made throughout its heliotropic movements.
The hormone ‘Auxin’ is vital for sunflower growth and stem elongation. It is
responsible for the alteration in natural movement of sunflowers and has a significant
impact on the heliotropism of sunflowers during the day. The growth hormone that is
activated throughout the usual hours of the day (24 h day/night cycle) is represented
by the parameter ‘Aux’ in SFOA. When the day’s hours are longer than the 24 h
day/night cycle, Auxin is not activated. The value of ‘Aux’ in this work was picked
at random from the range of [0, 1].
The heliotropic movements of immature sunflowers were regulated by a biological
clock. This clock measures the time it takes to complete one cycle of a 24 h day/night
cycle. The immature sunflowers did not migrate back and forth on a regular timetable
if the period of their heliotropic movements increased over the 24 h day/night cycle.
The parameter ‘Hours day’ in SFOA offers a random hour that is chosen as a random
integer value between [0, 100]. In Eq. (3), this parameter is employed to toggle
between the activation and deactivation states of the growth hormone ‘Auxin’.
Immature sunflowers grow in reaction to a movement known as heliotropism, as
previously stated. Heliotropism has been shown to be controlled not only by direct
light but also by their biological clock. On a cloudy or rainy day, the heliotropic
movements of immature sunflowers are slower [23]. This means that on a cloudy
or wet day, the Auxin is not triggered. The second mode has been used to replicate
this circumstance in the proposed SFOA, where the ‘Sun’ parameter set to ‘0’. The
following is the equation for this case:
Itr
tr
It
L Itr+1
new,SF = L old,SF + d × sin(ω) × L best,SF − L old,SF , at any hours day
(4)
To confirm that the ‘Auxin’ hormone is not activated on a cloudy or rainy day, the
influence of the ‘Aux’ parameter has been removed from this equation. An immature
sunflower can update its stem length according to random heliotropic motions in
response to accumulating the growth hormone ‘Auxin’ in one side without the other,
using Eq. (3) for sunny cases or Eq. (5) for overcast or rainy cases. As a result, by
shifting the initial sine function by phase angle ‘φ’, which is randomly chosen in
the range [0,0.01], the same concept can be further extended with Dim-dimensional
search space. Because the immature sunflower reorients itself towards the east at
night and west during the day, this cyclic prototype allows an immature sunflower to
be relocated around a different solution. This ensures that the space’s intensification
is specified as a choice between two solutions. Immature sunflower stems must be
232
Z. Qadir et al.
able to seek outside the local space of their corresponding best solutions in order to
broaden the search space. This can be accomplished by adjusting the Auxin’s range
during the course of a 24 h day/night cycle.
Naturally, the immature sunflower begins swaying with a little amplitude on its
own in pursuit of light. The oscillation’s tiny amplitude then increases and changes
over a considerably wider scale area. To calculate the promising areas of the search
space and eventually converge to the global optimum, stochastic optimization algorithms should be able to balance between diversification and intensification phases.
The range of the immature sunflower’s heliotropic movements is adaptively limited
during iterations to balance diversification and intensification. When the Auxin
is engaged and the sunflower’s biological clock is working normally, the SFOA
traverses the search space. When the Auxin is not engaged, or when the biological
clock is disrupted, this algorithm uses the search space to find anomalous heliotropic
movements. It means that the range of heliotropic movements should be reduced by
the sunflower’s stem.
When the enumerator exceeds the maximum number of iterations, the SFOA
completes the optimization process. Other stopping criteria, such as the maximum
number of fitness functions evaluated or the correctness of the calculated global
optimum, can be employed instead. The SFOA optimization procedure begins with
a random population of sunflower stem lengths (candidate solutions). Where the best
sunflower stem length is found, the values of the objective function are assessed. The
new population for the following iteration is determined using Eq. (3) for sunny mode
or Eq. (5) for cloudy mode, and the values of the objective function of all sunflowers
are assessed. The current best sunflower’s stem length will be compared against the
freshly created population’s best sunflower’s stem length, with the better one being
saved for the next iteration. This algorithm has been run until it meets a termination
requirement. The SFOA flowchart is shown in Fig. 2.
3.1 Problem Statement
The global optimum finding is the optimization problem that contains bounded search
space and to design a vector x such that the cost function f(x) minimizes. Path planning
is also the optimization problem in which path is needed to be minimized between
the start and endpoint.
First, important thing is to design a vector of control points that needs to be added
between the start and endpoint. Since every control point in x is connected with its
predecessor and successor point in the planning path as shown in Fig. 3. In this study
three control points are selected between the start and endpoint.
Since we can tune the no of control points the vector x containing control points
is X = (P1, P2, P3) T. The second important part is the cost or loss function f(x)
which needs to be minimized during the optimization. This function is nothing but
the Euclidean distance of a point from Ps to Pe which is starting and endpoint
respectively:
Optimizing UAV Path for Disaster Management in Smart Cities …
233
Fig. 2 Flow chart of proposed technique for UAV path optimization
Fig. 3 a Representation of workspace with blue circles as obstacles and start/end points b Path
optimized with defined control points
234
Z. Qadir et al.
n=2
||P L−−Pe + 1||) + ||Pn + Pe||
F(x) = ||Ps−−P1|| +
(5)
l=0
3.2 Path Optimization Using SFOA
SFOA will initialize control points randomly in the search space between Ps and
Pe. Then calculate the cost function and update the positions of control points. After
calculating the position of control points, the algorithm also checks for obstacle
violation (OV). This means the control point shouldn’t be at the position and in the
radius of the obstacle. This will avoid collision of the UAV with obstacles.
As shown in the Fig. 4 the shortest path in the workspace between Ps and Pe is
Path—1. But this will lead to the collision of the UAV with obstacles. Path—2 is also
the collision-free path but with a large distance. So, SFOA will extract the control
point positions such that the path should be less as shown in Path—3. Since Path—3
is generated very close to the boundary of the obstacle. Figure 4 shows the working
of SFOA for Path planning over the iterations in which control points are adjusted
with the best global cost and at the end of termination criteria the minimum path is
designed for UAV by SFOA.
Fig. 4 Multiple possible paths in workspace from start to end point
Optimizing UAV Path for Disaster Management in Smart Cities …
235
4 Case Studies with Discussion
In this section, four different scenarios are implemented and discussed. These
scenarios represent the real-time simulation of different problems. Different algorithms are implemented to test the performance. SFOA is compared with GHO,
FPA, and PSO. All algorithms are applied to test path optimization performance.
The hardware used for implementation is AMD PRO A9600B RS with 8 GB RAM
on MATLAB 2018a. The comparison factors used are the optimized cost and run
time required for the execution. The superior performance of SFOA can be observed
in results and discussion.
4.1 Scenario 1: General Environment
In this case 1 general environment is stimulated which contains a small number of
obstacles in the search space with different sizes and different positions. For UAV path
planning optimized path is one of the important factors which needs to be minimized
by the algorithm, SFOA’s dynamic grouping method makes it very effective for the
optimization and finding global best. High exploration at the start of iterations makes
it very suitable for path optimization. The Fig. 5 shows the comparison of best cost
vs iterations achieved by different algorithms. SFOA achieves less cost as compared
to other techniques. As shown in the Table 2, the cost achieved by SFOA, GHO, FPA,
and PSO is 10.6883 km, 12.1874 km, 12.3890 km, and 12.9814 km respectively.
Fig. 5 Cost versus iterations comparison of techniques in case 1
236
Table 2 Comparison of
techniques for case 1: GE
Z. Qadir et al.
Technique
Cost (km)
Run time (s)
SFOA
10.6883
147.96
GHO
12.1874
175.24
FPA
12.3890
167.44
PSO
12.9814
197.45
Fig. 6 Comparison of optimized path calculated by all techniques for UAV; a SFOA; b GHO; c
FPA; d PSO
The complexity of the implementation of different algorithms depends upon
the execution time required. The execution time of SFOA, GHO, FPA and PSO
is 147.96 s, 175.24 s, 167.44 s, 197.45 s respectively as shown in Fig. 6. So, the
superior performance of SFOA can be verified in case 1.
4.2 Scenario 2: Condense Obstacle Environment
This scenario is implemented to mimic the behavior of condensily developed in urban
areas which have highly dense buildings. So, UAV path planning is very effective for
dense areas. Obstacles with fixed radius and at different locations mimic the behavior
of different buildings in a well-planned society. Random numbers in the position
updation in PSO don’t make it effective for the complex optimization problem,
trapping into local maxima by PSO can be observed in Fig. 7.
Some problems faced by FPA are due to restricted movement of particles with
decreasing value of “a”. SFOA outperforms other techniques in this complex
problem. The Fig. 8 shows that SFOA achieves less cost over the iterations as
Optimizing UAV Path for Disaster Management in Smart Cities …
237
Fig. 7 Cost versus iterations comparison for case 2 for techniques
Fig. 8 Comparison of optimized path calculated by all techniques for UAV in case 2; a SFOA; b
GHO; c FPA; d PSO
compared to GHO, FPA, and PSO which is 10.179 km, 10.2107 km, 10.234 km
and 10.6207 km respectively. This also shows that SFOA continuously updates the
positions of particles according to the global best which doesn’t make it stuck at
LM. So, SFOA performs efficiently for Path planning in Scenario 2 also as shown in
Table 3.
238
Table 3 Comparison of
techniques for case 2: COE
Z. Qadir et al.
Technique
Cost (km)
Run time (s)
SFOA
10.1789
97.25
GHO
10.210
101.24
FPA
10.234
94.732
PSO
10.621
128.469
Complexity analysis is also done using the execution time. Time taken by SFOA,
GHO, FPA, and PSO is 97.251 s, 101. 240 s, 94.73 s and 128.469 s respectively.
Above discussed results shows that the performance of different techniques is SFOA
< GHO < FPA < PSO.
4.3 Scenario 3: Maze Environment
In this scenario, complex maze environment is presented to check the performance
of UAV path optimization techniques. This case is presented so that intelligent UAV
can effectively solve the maze with shortest path. Circle object are placed in such a
way that they behave like a wall of maze.
In this complex problem, every other algorithm except SFOA falls into local
optimum trap. As shown in the Fig. 9, the best cost achieved by SFOA is 12.995 km
and cost achieved by GHO, FPA, and PSO is 13.124 km, 13.226 km, 20.579 km, and
respectively. In specific iteration SFOA effectively optimize the UAV path.
Due to random oscillations in PSO and parameter decrement in FPA, both couldn’t
find the global optimum point. Dynamic grouping with variable population capability,
Fig. 9 Cost versus iterations comparison for all techniques in case 3
Optimizing UAV Path for Disaster Management in Smart Cities …
Table 4 Comparison of
techniques for case 3: ME
239
Technique
Cost (km)
Run time (s)
SFOA
12.995
124.75
GHO
13.124
131.89
FPA
13.226
134.21
PSO
20.579
140.92
Fig. 10 Comparison of optimized path calculated by all techniques for UAV in case 3; a SFOA; b
GHO; c FPA; d PSO
SFOA effectively minimize the cost and makes the UAV more intelligent for optimized path finding. This property of SFOA doesn’t make it stuck at local minima. The
updation of the particles is very effective in SFOA. For this scenario, the time taken
by all techniques i.e. SFOA, GHO, FPA and, PSO is 124.75 s, 131.89 s, 134.21 s,
140.90 s, respectively as shown in Table 4. Less time taken for the path optimization
in case 3 validates that the complexity of proposed technique is far more less than
other techniques. Performance comparison of all techniques can be made as SFOA
< GHO < FPA < PSO as shown in Fig. 10.
4.4 Scenario 4: Dynamic Environment
In this paper dynamic environment i.e. mixture of case 2 and case 3 is presented which
contains maze type environment with distributed objects. This complex scenario will
help UAV to find a path with more intelligent way. Distributed obstacles provide a
240
Z. Qadir et al.
Fig. 11 Cost versus iterations for all techniques in case 4
different environment, it helps to make UAV that much intelligent that is should also
look for the obstacles during Path finding.
Again, in this complex environment, PSO, FPA, stuck into local maxima and
couldn’t find shortest path for the UAV. This shows that the other algorithms are not
best for the UAV path optimization in such complex environment.
In this case as shown in the Figs. 11 and 12 the optimized path calculated by
SFOA, GHO, FPA, PSO is 10.9719, 13.2889, 13.668, 14.547 with time taken is
109.42 s, 113.568 s, 115.250 s, 126.210 s respectively as shown in Table 5.
4.5 Performance Evaluation
In this study, comparison is made between PSO, FPA, GHO and SFOA for
UAV path optimization capability. To test this, four different cases are presented
which includes general environment, condense environment, maze environment and
dynamic environment.
As discussed previously, dynamic grouping in SFOA with both exploration and
exploitation groups working simultaneously makes it effective for UAV path planning. Parameters used to guage the performance of comparative techniques are optimized path length and the execution time with population size 100, no of iterations
100 and 3 control points the SFOA achieves up to 24.5% less cost as compared
to other techniques with up to 13.3% less time taken. So, SFOA can be efficiently
applied to optimize path of UAV.
Optimizing UAV Path for Disaster Management in Smart Cities …
241
Fig. 12 Comparison of optimized path calculated by all techniques for UAV in case 4; a SFOA; b
GHO; c FPA; d PSO
Table 5 Comparison of
techniques for case 4: DE
Technique
Cost (km)
Run time (s)
SFOA
10.9719
109.42
GHO
13.2889
113.568
FPA
13.892
119.191
PSO
14.547
126.210
5 Conclusion
In this study, various state-of-the-art metaheuristic algorithms are tested in order
to give a collision-free and time-efficient path planning for UAVs in a catastrophic
event. SFOA, PSO, FPA, and GHO are among the algorithms used. Furthermore, the
usefulness of these algorithms is evaluated not only in a static environment, but also
in four other scenarios: general, condensed, maze, and dynamic environments. In
all environments, the exploration and exploitation groups are integrated with a 70–
30 strategy that works simultaneously for global optimal and global best solutions.
The results suggested that SFOA algorithm beats existing methods, achieving up to
24.5% lower transportation costs and up to 13.3% lower computing time.
242
Z. Qadir et al.
References
1. H. Liu, X. Li, M. Fan, G. Wu, W. Pedrycz, P.N. Suganthan, An autonomous path planning
method for unmanned aerial vehicle based on a tangent intersection and target guidance strategy.
IEEE Trans. Intell. Transp. Syst. 1–13 (Nov 2020). https://doi.org/10.1109/TITS.2020.3030444
2. J. Li, Y. Xiong, J. She, M. Wu, A path planning method for sweep coverage with multiple
UAVs. IEEE Internet Things J. 7(9), 8967–8978 (2020). https://doi.org/10.1109/JIOT.2020.
2999083
3. X. Zhou, F. Gao, X. Fang, Z. Lan, Improved bat algorithm for UAV path planning in threedimensional space. IEEE Access 9, 20100–20116 (2021). https://doi.org/10.1109/ACCESS.
2021.3054179
4. L. Shi, Z. Jiang, S. Xu, Throughput-aware path planning for UAVs in D2D 5G networks. Ad
Hoc Netw. 116 (2021). https://doi.org/10.1016/j.adhoc.2021.102427
5. R. Masroor, M. Naeem, W. Ejaz, Efficient deployment of UAVs for disaster management: A
multi-criterion optimization approach. Comput. Commun. 177, 185–194 (2021). https://doi.
org/10.1016/j.comcom.2021.07.006
6. N.A. Kyriakakis, M. Marinaki, N. Matsatsinis, Y. Marinakis, Moving peak drone search
problem: An online multi-swarm intelligence approach for UAV search operations. Swarm
Evol. Comput. 66 (2021). https://doi.org/10.1016/j.swevo.2021.100956
7. S. Wan, J. Lu, P. Fan, K.B. Letaief, Toward big data processing in IoT: Path planning and
resource management of UAV base stations in mobile-edge computing system. IEEE Internet
Things J. 7(7), 5995–6009 (2020). https://doi.org/10.1109/JIOT.2019.2954825
8. C. Tabasso, N. Mimmo, V. Cichella, L. Marconi, Optimal motion planning for localization of
avalanche victims by multiple UAVs. IEEE Control Syst. Lett. 5(6), 2054–2059 (2021). https://
doi.org/10.1109/LCSYS.2021.3049314
9. M. Onishi, T. Ise, Explainable identification and mapping of trees using UAV RGB image and
deep learning. Sci. Rep. 11 (1) (2021). https://doi.org/10.1038/s41598-020-79653-9.
10. G. Castellanos, G. Vallero, M. Deruyck, L. Martens, M. Meo, W. Joseph, Evaluation of flying
caching servers in UAV-BS based realistic environment. Veh. Commun. 32 (2021). https://doi.
org/10.1016/j.vehcom.2021.100390
11. J. Liu, X. Qin, B. Qi, X. Cui, 3D online path planning of UAV based on improved differential
evolution and model predictive control. Int. J. Innov. Comput. Inf. Control 16(1), 315–329
(2020). https://doi.org/10.24507/ijicic.16.01.315
12. Z. Huang, C. Chen, M. Pan, Multiobjective UAV path planning for emergency information
collection and transmission. IEEE Internet Things J. 7(8), 6993–7009 (2020). https://doi.org/
10.1109/JIOT.2020.2979521
13. S. Zhu, L. Gui, N. Cheng, F. Sun, Q. Zhang, Joint design of access point selection and path
planning for UAV-assisted cellular networks. IEEE Internet Things J. 7(1), 220–233 (2020).
https://doi.org/10.1109/JIOT.2019.2947718
14. S.-H. Kim, G.E.G. Padilla, K.-J. Kim, K.-H. Yu, Flight path planning for a solar powered UAV
in wind fields using direct collocation. IEEE Trans. Aerosp. Electron. Syst. 56(2), 1094–1105
(2020). https://doi.org/10.1109/TAES.2019.2926654
15. N.-N. Dao et al., Survey on aerial radio access networks: Toward a comprehensive 6G access
infrastructure. Accessed: 02 Mar 2021. [Online]. Available: https://www.6gchannel.com
16. J.-A. Delamer, Y. Watanabe, C.P.C. Chanel, Safe path planning for UAV urban operation under
GNSS signal occlusion risk. Rob. Auton. Syst. 142 (2021). https://doi.org/10.1016/j.robot.
2021.103800
17. C. Xu, M. Xu, C. Yin, Optimized multi-UAV cooperative path planning under the complex
confrontation environment. Comput. Commun. 162, 196–203 (2020). https://doi.org/10.1016/
j.comcom.2020.04.050
18. C. Duan, J. Feng, H. Chang, Meteorology-aware path planning for the UAV based on the
improved intelligent water drops algorithm. IEEE Access 9, 49844–49856 (2021). https://doi.
org/10.1109/ACCESS.2021.3068972
Optimizing UAV Path for Disaster Management in Smart Cities …
243
19. S. Zhang, R. Zhang, Radio map-based 3D path planning for cellular-connected UAV. IEEE
Trans. Wirel. Commun. 20(3), 1975–1989 (2021). https://doi.org/10.1109/TWC.2020.3037916
20. G.E.G. Padilla, K.-J. Kim, S.-H. Park, K.-H. Yu, Flight path planning of solar-powered UAV for
sustainable communication relay. IEEE Robot. Autom. Lett. 5(4), 6772–6779 (2020). https://
doi.org/10.1109/LRA.2020.3014635
21. H. Moayedi, H. Nguyen, L. Kok Foong, Nonlinear evolutionary swarm intelligence of
grasshopper optimization algorithm and gray wolf optimization for weight adjustment of neural
network. Eng. Comput. 37(2), 1265–1275 (2021)
22. X.S. Yang, M. Karamanoglu, Nature-inspired computation and swarm intelligence: a state-ofthe-art overview. Nat. Inspired Comput. Swarm Intell. 3–18 (2020)
23. D. Sattar, R. Salim, A smart metaheuristic algorithm for solving engineering problems. Eng.
Comput. 37(3), 2389–2417 (2021). https://doi.org/10.1007/s00366-020-00951-x
24. S. Mirjalili, S.M. Mirjalili, A. Lewis, Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61 (2014).
https://doi.org/10.1016/j.advengsoft.2013.12.007
25. J. Chen, C. Du, Y. Zhang, P. Han, W. Wei, A clustering-based coverage path planning method
for autonomous heterogeneous UAVs. IEEE Trans. Intell. Transp. Syst. (2021). https://doi.org/
10.1109/TITS.2021.3066240
26. A. Kaveh, Advances in metaheuristic algorithms for optimal design of structures (Springer
International Publishing, Switzerland, 2021), pp. 9–40
27. N. Arora, R.K. Banyal, Workflow scheduling using particle swarm optimization and gray wolf
optimization algorithm in cloud computing. Concurrency Comput. Pract. Experience e6281
(2021)
28. A. Köybaşi, İ. Yazici, Solution of test problems with grey wolf optimization algorithm and
comparison with particle swarm optimization. Sakarya Üniversitesi Fen Bilimleri Enstitüsü
Dergisi 24(6), 1252–1264 (2020)
29. M.H. Mansoor, A.F. Mirza, Q. Ling, M.Y. Javed, Novel grass hopper optimization based MPPT
of PV systems for complex partial shading conditions. Sol. Energy 198, 499–518 (2020)
30. Z. Qadir, F. Al-Turjman, M.A. Khan, T. Nesimoglu, ZIGBEE based time and energy efficient
smart parking system using IOT, in 2018 18th Mediterranean Microwave Symposium (MMS)
(IEEE), pp. 295–298
31. Z. Qadir, S.I. Khan, E. Khalaji, H.S. Munawar, F. Al-Turjman, M.P. Mahmud, K. Le, Predicting
the energy output of hybrid PV–wind renewable energy system using feature selection technique
for smart grids. Energy Rep.
32. S.I. Khan, Z. Qadir, H.S. Munawar, S.R. Nayak, A.K. Budati, K.D. Verma, D. Prakash, UAVs
path planning architecture for effective medical emergency response in future networks. Phys.
Commun. 47, 101337 (2021)
33. Z. Qadir, F. Ullah, H.S. Munawar, F. Al-Turjman, Addressing disasters in smart cities through
UAVs path planning and 5G communications: A systematic review. Comput. Commun. (2021)
34. Z. Qadir, E. Ever, C. Batunlu, Use of neural network based prediction algorithms for powering
up smart portable accessories. Neural Process. Lett. 53(1), 721–756 (2021)
35. M. Atif, R. Ahmad, W. Ahmad, L. Zhao, J.J.P.C. Rodrigues, UAV-assisted wireless localization
for search and rescue. IEEE Syst. J. (2021). https://doi.org/10.1109/jsyst.2020.3041573
36. N. Wen, L. Zhao, X. Su, P. Ma, UAV online path planning algorithm in a low altitude dangerous
environment. IEEE/CAA J. Autom. Sin. 2(2), 173–185 (2015). https://doi.org/10.1109/JAS.
2015.7081657
37. M.N.A. Wahab, C.M. Lee, M.F. Akbar, F.H. Hassan, Path planning for mobile robot navigation
in unknown indoor environments using hybrid PSOFS algorithm. IEEE Access 8, 161805–
161815 (2020). https://doi.org/10.1109/ACCESS.2020.3021605
38. F. Gul, I. Mir, W. Rahiman, T.U. Islam, Novel implementation of multi-robot space exploration utilizing coordinated multi-robot exploration and frequency modified whale optimization algorithm. IEEE Access 9, 22774–22787 (2021). https://doi.org/10.1109/ACCESS.2021.
3055852
39. A.V. Savkin, H. Huang, Navigation of a network of aerial drones for monitoring a frontier of
a moving environmental disaster area. IEEE Syst. J. 14(4), 4746–4749 (2020). https://doi.org/
10.1109/JSYST.2020.2966779
244
Z. Qadir et al.
40. H.H. Cho, H. Te Wu, C.F. Lai, T.K. Shih, F.H. Tseng, Intelligent charging path planning for
IoT network over blockchain-based edge architecture. IEEE Internet Things J. 8(4), 2379–2394
(2021). https://doi.org/10.1109/JIOT.2020.3027418
41. N.B. Rossello, R.F. Carpio, A. Gasparri, E. Garone, Information-driven path planning for UAV
with limited autonomy in large-scale field monitoring. IEEE Trans. Autom. Sci. Eng. 1–11
(June 2020). Accessed: 17 Aug 2021. [Online]. Available: https://arxiv.org/abs/2006.11000v3
42. M.A.A. Khan, Z. Qadir, M. Asad, A.Z. Kouzani, M.A. Parvez Mahmud, Environmental footprint assessment of a cleanup at hypothetical contaminated site. Appl. Sci. 11, 4907 (2021).
https://doi.org/10.3390/app11114907
43. A.B. Ajmal, M. Alam, A.A. Khaliq, S. Khan, Z. Qadir, M.A.P. Mahmud, Last line of defense:
reliability through inducing cyber threat hunting with deception in SCADA networks. IEEE
Access 9, 126789–126800 (2021). https://doi.org/10.1109/ACCESS.2021.3111420
44. M.D. Phung, Q.P. Ha, Motion-encoded particle swarm optimization for moving target search
using UAVs. Appl. Soft Comput. 97, 106705 (2020)
UAV-Based Rescue System and Seismic
Zonation for Hazard Analysis
and Disaster Management
Abdul Qahar Shahzad, Mona Lisa, Mumtaz Ali Khan, and Irum Khan
Abstract The rapid growth of population leads to numerous difficulties in case of
any natural calamity. Such problems include insecurity, risk assessment and disaster
planning of affected regions. Therefore, the critical demand of society is to invent
efficient ways to tackle disastrous situations. UAV-based technologies are the only
realistic approach to deal with such circumstances. Additionally, UAVs are already
acknowledged for low cost-effectiveness and flexibility by executing numerous tasks.
This work proposes an autonomous system based on drone rescue (Auto-SBDR).
Auto-SBDR model uses UAVs to provide first and quick response to a life-threatening
situation in urban areas like residential and commercial setups. However, integrated
autonomous rescue systems comprise three main sections: intelligent UAVs, sensor
networks and command centers. Furthermore, Auto-SBDR systems are designed
in such a manner that UAVs automatically deploy to a specified location based on
seismic zonation upon receiving signals from the sensors network. Where, seismic
zones are produced through computing earthquake catalogs. Through Interpolation
techniques, Kalabagh areas are categorized in events frequency, depth and magnitude.
Moreover, the projected UAV rescue system drastically reduces response time as
compared to conventional response systems. While operating in earthquake-prone
areas, data collected by FANETs are considered to be extremely valuable.
Keywords FANETs · UAV · Auto-SBDR · Seismic zonation · Earthquake
A. Q. Shahzad (B) · M. Lisa
Department of Earth Sciences, Quaid-I-Azam University, Islamabad 45320, Pakistan
e-mail: aqshahzad@geo.qau.edu.pk
M. Lisa
e-mail: mlisa@qau.edu.pk
M. A. Khan · I. Khan
Department of Earth and Environmental Sciences, Bahria University, Islamabad 75260, Pakistan
e-mail: mumtazkhan@bui.edu.pk
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_14
245
246
A. Q. Shahzad et al.
1 Introduction
Throughout the globe, escalating number of disasters whether natural or man-man
hit numerous areas in recent years. Especially, most hazardous and predominant
disasters are earthquakes which affect millions of people and bring enduring damage
to active tectonic places. Therefore, to sustain operation in such area it is necessary
to initiate a plan which covers the basic needs of the affected people. It is necessary
to design an approach through which the victims are served in a better way such as
supplying food, medical equips, relief material, sharing information, and building
a communication system. Particularly, in such post-seismic atmosphere provision
for the social order is a benevolent arrangement where it compromises the material shipping, maintaining the mode of communication, rescuing victims from a bad
situation, and medicinal assistances from rescue base station (commonly known as
point of origin) to affected areas (also known as recipients’ site). There are numerous
witnessed catastrophes in the current century, some of them are followed as in 2011
the Haitian earthquake, in the year 2004 tidal wave in the Indian-Ocean, 2005 7.6
magnitudes of the earthquake in Pakistan northern areas, 2013 Pak-India flooding
or 2011 alarming crisis in Africa, specify that all over the world frequently developing nations are highly susceptible to destructive phenomena’s. Apart from this,
there are thousands of such crises and cases related to natural disasters which are
not properly highlighted but in terms of impact, they have the same catastrophic
power to affect the areas. However, the factor of urbanization as the escalating
trend of population and reduction in land per person lead to enhance the people
ratio in the unsafe areas. Consequently, an increasing number of people in such
disaster-prone area leads to greater number of casualties and elevate the victim as
earthquake strike such area [1]. Especially, in developing nation’s impairments of
altruistic assistance are strengthened by wellbeing amenities failure, the collapse
of already existing health and treatment in emergency cases, entire health facilities
disturbance, shortage of staff and service while catastrophe. Additionally, meager
sanitation and unclean water circumstances joint with low-vaccination treatment
repeatedly lead to numerous borne infections (air and water-borne diseases), which
include mainly malaria, measles, diarrheal diseases, acute-respiratory-infections,
leptospirosis, dengue fever, typhoid, viral-hepatitis, meningitis, as well as tetanuscutaneous [2]. In such post-seismic circumstances, quick-response along with rapiddistribution of energetic assistance substances which mainly includes RTUF (ready
to use therapeutic food) parcels, medical kits, H2 O decontamination pills, and injection into victims of the disaster-prone area that could protect lives or even slow
down the spread of the epidemics. Thus, colossal issues along with the delivery challenges of release matters in emerging nations are also connected with the specified
modes of transportation and shipping setup. Furthermore, non-government organization automobile means in emerging states are fairly inadequate plus very expensive
due to mounting gasoline ingestion, assurance, and maintenance cost. Growth along
with the disaster tasks is usually accompanied through old automobile convoys as
postponed truck replacement is away from the suggested time edge. Poor outdated
UAV-Based Rescue System and Seismic Zonation …
247
accustomed truck group limits packing capabilities. Therefore, light-weighted materials are suggested due to its efficiency in performance. Land-based mechanical
carriage means through charitable administrations are usually restricted to SUVs
(Sport-utility-vehicles) and automobiles of small size. The main reason behind this
approach is the severity of the situation as greater mode of transportation is not viable
in such hazardous conditions [3]. Additionally, in such situation relief organizations
are usually faced with outdated or improper road and other infrastructure. Whereas,
carpeted road worth in case of emerging nations is categorized by width, proportion
of paved street and material used during construction of road. Terrestrial physical
appearance, for instance geologically disseminated topography, mountainous region,
and islands, signify supplementary impairments to existing serious condition [4].
While critical and life-threatening disastrous, existing outdated situation become
more frightening as water channel floods, road disrupt, bridge collapse, communication tools disables, landslides occur which block the transportation, open area
covers with debris, fire spread due to exposure of volatile substance, and sudden or
catastrophic earthquake seismic waves fluctuate the real identity of an area through
variation in both layer air/land. [5, 6]
Main objectives of this study is to provide disaster-managers with outcomes
exposed through evaluation of 3-D assessed model along with plotting for possible
extra importance which carry airborne valuations of earthquake associated destruction and requirement. However, the approach of 3D assessment model is already
proved and illustrated its capabilities in diverse field, for example construction
model. However, infrastructures are not design on the disaster response nor there
is any proper guideline for construction. Therefore, urgent requirement to assessed
3D design model and plans for calamity airborne surveys are conducted in order to
design a proper guideline for disaster prone humanities’ [7]. Through assessment it is
understood to accomplished comparable facts in added spheres, whereas proposition
expressed for investigation adopts the 3-dimensional and plotting (orthophoto) will
develop airborne calculations of earthquakes-related necessities and recompenses
[8–10]. Though suggestion it is completely authenticated by statistic that particular
3D design allows users to appreciate construction’s altitudes from diverse viewpoints, where it is entirely impossible through 2D angle. Moreover, through this
utilized features especially when designing a structure, it is strongly recommended
that exploration related to disaster is proved more efficient. In order to mitigate this
issue zonation is performed before describing the construction guideline for any
building sites in case of earthquake prone area. Along with micro size zonation, the
UAV based analysis is conducted to improve overall performance while rescuing the
victims’ [11].
2 Tectonic Setting of Kalabagh Area
Tectonic-based studies are conducted in the vicinity of Kalabagh fault where two
prominent faults and ranges exist as given below.
248
A. Q. Shahzad et al.
2.1 Salt Range and Trans Indus Range Thrust
The (SRT) Salt Ranges and Trans Indus Range Thrust (TRIST) lies in the
extreme south-most part of northern Pakistan Himalayan thrust system. Where,
mentioned thrust-fault describes the active-front deformational range alongside
with the Cambrian. Additionally, rock of Paleocene rocks is completely thrusted
into the undeformed Indus Plain sediments. Therefore, the existing day performed
as a basin-shaped for the weather-beaten substantial of debris in the vicinity of
Northern-Himalayan mountainous chain [12].
2.2 Surghar Fault
Furthermore, this particular mountainous range which is known as Surghar Range
lies in the Trans-Indus-Range where it illustrates the southern deformation of the
Himalayan-Orogeny. Additionally, besides the Surghar-Range-Thrust, especially the
Mesozoic sediments are completely thrusted above the southern Punjab-Plain as
shown in Fig. 1. Therefore, this range is continuously moving in south-ward direction. Main reason of this tectonic movement is thrusting of basement (Indian-plate)
beneath the sedimentary rocks cover [13, 14]. However, physical geology along with
Fig. 1 This maps shows study area Kalabagh fault zone on satellite imagery using digital elevation
model (DEM)
UAV-Based Rescue System and Seismic Zonation …
249
the geometrical characteristics of the Surghar-Range is due to the vicinity of ChichaliNala extent. Where, it is further characterized through the anticline (verging in the
south). Moreover, Surghar-Anticline lies at the Jurassic basement. Thus, it is understood that the fault-bend-fold lies above the local is disconnected. This separated
layer lies in the basement of Triassic age rocks. Beside this along the Surghar-Fault
the mountain is entirely thrusted in the South direction of Punjab-foreland, where it
is striking as South-verging-Fore-thrust. Hence, it represents a continuing restricted
and defused geometry along the Surghar-Fault striking. Due to these features this
fault-line is considered as extremely growing thrust in the west direction (in the
vicinity of Kutki village) where it have a very deep thrust-sheet [15, 16]. The above
mentioned featured of the this specific fault along with aerial survey conclude that
that due to reinstated structural cross-section in the vicinity of Kutki-village-crosssection revealed that Surghar-fault reduce to approximately 5.6 km [17, 18]. While
the distortion proposes that the approximately timing for uplifting along this range
is initiated 2.3 Million years ago this is situated beneath the thrust area [19].
3 Approaches for Seismic Zonation of Kalabagh
For seismic zonation, there are numerous interpolation techniques. In this chapter,
two renowned image-based interpolations approaches are used for higher visibility.
3.1 Kriging Methodology
Kriging is the practice of interpolations which mainly focus on the computation
of anonymous point or statistics from perceived information at familiar positions.
However, in mathematical-based model which is formerly in geostatistics. Additionally, Gaussian-method or Kriging-regression is interpolation-based techniques
where inserted information is visualized through appropriate prior covariance [20].
Statistically, Kriging-interpolation is accessible through following equation.
Z 0∗ =
n
λi0 Z i
(1)
i=1
Here the parameter Z∗0 symbolizes the anonymous information. Whereas, Zi
represents the understood statistics in image processing. The factor λ0i is used to
identify it weight. Kriging mathematical-approach particularizes the assortment of
appropriate weights. Although procedure performs avariogram to explain accurate
3-D (three-dimensional), it condenses inaccuracy of the predictable data that are
predictable through the spatial-spreading of identified points [21].
250
A. Q. Shahzad et al.
Therefore, broad-spectrum linear-based leaning design is fashioned through
universal Kriging practice. This approach mainly consists of drift-functions (represented as numerical) to analyze m ∗ x, which is considered as possibility of
z ∗ x.
Assume
m(x) = a0 + a1 u + a2 v + a3 u2 + a4 uv + a5 v2
(2)
where, u and v is route or coordinate of value (x)? Through this approach the achieved
equation is followed as;
λ0i a0 + a1 xi + a2 yi + a3 x2i + a4 xi yi + a5 y2i
i
= a0 + a1 x0 + a2 y0 + a3 x20 + a4 x0 y0 + a5 y20
(3)
Through altering the mathematical Eq. 3, subsequent calculation is attained
⎧ 0
0
⎨ i λi = 1;
i λi xi = x0
0
λ
y
=
y
;
λ0 x2 = x20 ;
0
i i0 i2
⎩ i i0 i
2
i λi xi yi = x0 y0 ;
i λi yi = y0
(4)
Set
λ0i Pl (xi ) = Pl (x0 ), (l = 0, 1, 2, 3, 4, 5),
(5)
i
where Pl = 1, x, y, x2 , xy, y2 .
2
= Var(Z0 ) +
λ0i λ0j c xi xj − 2
λ0i c(xi x0 ),
E Z∗0 − Z0
i
j
(6)
i
where c xi xj = COV Zi Zj and c(xi x0 ) = COV(Zi Z0 ), constructed or shaped
through utilizing the relationship of Lagrange multiplier (*) rule [22].
⎧
5
⎪
⎨ λ0i c xi xj − μl Pl (xi ) = c(xi x0 ) (i = 1, 2, · · · , n)
i
l=0
⎪
⎩ λ0i Pl (xi ) = Pl (x0 )
l = 0, 1 − −5
(7)
i
All adapted in matrix ‘[]’ preparation, likewise Ax = b analyze information of
λ0i (i = 1, 2, 3, . . . , n). Where the Step. 1, conclusion is attained approximation of
unidentified values [23].
UAV-Based Rescue System and Seismic Zonation …
3.1.1
251
Kalabagh Seismicity
The study is focused on the earthquake distribution and hazard analysis of Kalabagh
area. The area hosts a major strike slip fault called the Kalabagh strike slip fault. The
earthquake magnitudes are plotted and then interpolated over this region by using
Kriging interpolation techniques (e.g. Kriging by using nearest neighbor, Kriging
by using Cubic Convolution, Kriging by using Bilinear Interpolation) as shown in
Fig. 1. The area has a maximum recoded range of earthquake magnitude of greater
than 4.9 (> 4.9) and less than 6.5 (< 6.5) which lies in the moderate earthquake range
in terms of magnitude as shown in Fig. 2. Additionally, Fig. 3 demonstrates with
15% lucidity where Fig. 4 depicts nearest neighbor convolution.
3.2 Cubic Convolution
This technique particularly focuses on the special function of (x). However, ultimate
character of this approach is that it is agreed with data points. Beside this subordinate
in terms, if (x) is the function then f(x) is analogous to CC technique symbol. Where,
X is equal to X(xi). Though, the (xi) is the representation of interpolation-nodes [24].
Correspondingly, when there is adjacent spreading of points. The numerous sign
x is represented in the following way.
Fig. 2 Map shows the kriging on study area by using nearest neighbor method plotted on Rana
and Kazmi Geological map
252
A. Q. Shahzad et al.
Fig. 3 Map shows the kriging on study area 15% transparency by using nearest neighbor method
plotted on Rana and Kazmi Geological map
Fig. 4 Map shows the kriging on study area by using nearest neighbor method plotted on DEM
g(x) =
x − xk
ck u
h
(8)
UAV-Based Rescue System and Seismic Zonation …
253
where, (f ) is categorized in the middle of linear-series and Cubic-spline (Hou and
Andrews). h express the information improvement. Where, u is kernel interpolation.
Also, xk is node. The factor g is function. However, parameter ‘ck’ is formed through
points [26]. Through such arrangement the interpolation from f(xk ) = g(xk ) for every
xk , is figure out.
Furthermore, principal Kernel-technique is altered in numerous ways to unbreakable gx series through mathematical model of convolution. Additionally, Kerneltechnique has an impact on mathematics based functions. Therefore, due to strong
influences on impact-effectiveness, kernel is efficiently process to generate procedures. This approach resultant from assemblage circumstances computed on kernel.
Where, it is considered to improve accurateness level of mathematical determination
[25].
3.2.1
Mathematical Approach for Cubic Convolution
This approaches work on the subset in a particular interval (− 2 − 1, − 1 0, 0 1, 1 2).
Where, the subset outer interval range from ‘− 2’ to the positive ‘2’ and the kernel is
equal to zero. According to the significant situation, numerous values processed to
reach the process of CC. whereas, phase (1) is reduced to phase (4). The Kernel-based
approach is purely symmetrical calculation [26]. Additionally, the former joint term,
‘u’ is sustained in such condition.
⎧
3
2
⎨ A1 |s| + (C1 |s|) + (B1 |s| ) + (D1 ) 0 < s < (1)
3
3
(9)
u(s) = A2 |s| + C2 |s| + B2 |s| + D2
(1) < s < 2
⎩
0
(2) < s.
Kernel-technique is mainly utilized to figure out the u(n) = 0 and u(0) = 1.
When there are numerous non-zero integers. Where, this circumstance has strong
computational importance. ‘h’ is collecting boost. Main transformation amongst the
node xj is (+ j − k) * h. However, xj is supported in step (1). Therefore, the resultant
mathematical equation is followed as;
ck u(j − k).
g xj =
(10)
k
In case k = j the u*(+j − k) will be zero (0) onother
side
[27].
Minimize to c (j). Kernel circumstances acquire f xj = g xj . Thus, f xj is equal
to the cj. Through further processing the ck ’s the mathematical approach is altered
by dataset. The terms and conditions for u = 1 and u ∗ 1 = u ∗ 2 = 0 offer 4 step
the number [28].
1 = u(0) = D1
0 = u 1− = A1 + C1 + B1 + D1
254
A. Q. Shahzad et al.
0 = u 1+ = A2 + C2 + B2 + D2
0 = u 2− = 8A2 + C2 + 4B2 + D2
Here it illustrates a 3-D chain design model at nodes (0, 1, and 2).
−C1 = u 0− = u 0+ = C1
3A1 + 2B1 + C1 = u 1− = u 1+ = 3A2 + 2B2 + C2
12A2 + 4B2 + C2 = u 2− = u 2+ = 0.
Taylor series is utilized along with the Nfham and Bretson approaches to obtain
all these seven mathematical equations. Thus, for attaining it we assumes that ‘A2’
is equal to ‘a’ as A2 = a . However, continuing all these seven expressions, model
for (a) is formed before entire calculations of seven equations [29]. Where, for the
kernel-based solution, symbol a, will be
⎧
3
2
⎨ (a + 2)|s| − (a + 3)|s| + 10 < |s| < 1
3
2
u(s) = a|s| − 5a|s| + 8a|s| − 4a1 < |s| < 2 .
⎩
01 < |s|
(11)
Here we assume that ‘x’ is dataset which is gathered during interpolation technique
processing. In case ‘x lies between the two-serial-node then, it is represented as xj
and xj+1 .
x−x
Suppose s = h j .
x−xj +xj −xk
k
Hence x−x
=
= s + j − k, (1) is designed as
h
h
g(x) =
Ck u(s + j − k)
(12)
k
Furthermore, when the parameter u is equal to zero in the subinterval (− 2, 2)
then j is 0 < S < 1, (−2, 2j, and since 0 < s < 1) reduces to
g(x) = cj−1 u(s + 1) + cj u(s) + cj+1 u(s − 1) + cj+2 u(s − 2).
Form equation number 4, it follow that
u(s + 1) = a(s + 1)3 − 5a(s + 1)2 + 8a(s + 1) − 4a
= as3 − 2as2 + as
u(s) = (a + 2)s3 − (a + 3)s2 + 1
u(s − 1) = −(a + 2)(s − 1)3 − (a + 3)(s − 1)2 + 1
UAV-Based Rescue System and Seismic Zonation …
255
= −(a + 2)s3 + (2a + 3)s2 − as
u(s − 2) = a(s − 2)3 − 5a(s − 2)2 − 8a(s − 2) − 4
= as3 + as2
Here we divide the Eq. 6 correlation into fragments. Thus the sampling powers
(S), along with the CC sampling expression formed [30].
g(x) = − a cj+2 − cj1 + (a + 2) cj+2 − cj s3
+ 2a cj+1 − cj1 + 3 cj+1 − cj + a cj+2 − cj s2
− a cj+1 − cj−1 s + cj
where, the function f is three time series derivation in sub-interval [xj , xj+1 ], through
the approach of Taylor’s algorithm. The equation after derivation is followed as;
cj+1
f xj h2
+ 0 h3
= f xj+1 = f xj + f xj h +
2
(13)
Therefore, h = xj+1 − xj .0(h3 )’ represent the condition (h3 ). Where functions
reach to zero (0) at proportional to (h3 ).
This order is followed as;
cj+2 = f xj + 2hf xj + 2h2 f xj + 0 h3
cj−1
h2 f xj
+ 0 h3
= f xj + hf xj +
2
(14)
(15)
Further the equation eight and nine is integrated in equation number seven. The
mathematical-model equation for CC kernel-functions is achieved [30].
g(x) = −(2a + 1) 2hf xj + h2 f xj s3
(4a + 3)h2 f xj
+ (6a + 3)hf xj +
s2
2
− 2ahf xj s + f xj + 0 h3 .
(16)
Here sh = x − xj . Taylor algorithm for function x is;
2 2 f xj
+ 0 h3 .
f(x) = f xj + shf xj + s h
2
Now subtract the equation eleven from twelve.
(17)
256
3.2.2
A. Q. Shahzad et al.
f(x) − g(x) = (2a + 1) 2hf xj + h2 f xj s3
− (2a + 1) 3hf xj + h2 f xj s2
+ (2a + 1)shf xj + 0 h3 .
(18)
f(x) − g(x) = 0 h3 .
(19)
⎧
3
2
5
0 < |s| < 1
⎨ 3/2|s| − 2 |s| + 1
3
5
u(s) = −1/2|s| + 2 |s|2 − 4|s| + 2 1 < |s| < 2
⎩
0
2 < |s|.
(20)
Boundary Condition in Cubic Convolution
Initial exchange of function f has been experimented which define actual figures.
However, in more practical way f , is detected during restricted sub-intermission.
Therefore, domain off controlled to finite intervals calls (a, b) edge.
Complete model points xk initially must settle with the subset remarks (a, b).
Consider xk = xk−1 + h for k = 1, 2, 3, . . . N , where, x0 = a, xn = b and h = b−a
N
for N . However, number N should be selected through Nyquist-principle. Additionally, previous percentage is effective for any sort of set. Thus, homogenously dataset
point neither enhanced through newly formed classification for the x k ’s [25].
Where, Cubic Convolution interpolation function is expressed in form for interval
(a, b). The CC interpolation as;
g(x) =
N+1
k=−1
ck u
x − xk
h
(21)
While regulating function g for entire subset intermission x is (a, b). Though
values of ck is k = −1, 0, . . . , N + 1 which is acquired. Then the value for k along
with ck will be k = 0, 1, 2, N, ck = f(xk ).When k = −1 and for k = N + 1, still
the worth of function f is unknown. Since x−1 and xN+1 fall outside surveillance
intermission. Where, values assigned to c−1 and cN+1 are known as boundary set-up
[29].
However, boundary conditions are chosen so that g(x) is an 0 h3 approximation
to f(x) for entire function x confined to interim (a, b). Now consider applicable tools
for left-handed boundary. Suppose x is a point in the interim of [x0 , x1 ]. Due to this
factor the CC function reduced to the following equation [31].
g(x) = c−1 u(s + 1) + c0 u(s) + c1 u(s − 1) + c2 u(s − 2)
(22)
UAV-Based Rescue System and Seismic Zonation …
257
0
where s = x−x
, is obtained through integration of the mathematical formula number
h
fifteen for the function u. Through assembling the s power in order to further improve
the CC interpolation approach.
s3 c2 − 3c1 + 3c0 − c−1
g(x) =
2
s2 c2 − 4c1 + 5c0 − 2c−1
s c1 − c−1
−
(23)
+
+ c0 .
2
2
In case g is an 0 h3 estimate for f. Therefore, cubic power of s3 is equal to 0. It
concludes that the factor c−1 is added to improve the CC efficiency.
c−1 = c2 − 3c1 + 3c0 , or
c−1 = f(x2 ) − 3f(x1 ) + 3f(x0 ).
(24)
Here the mathematical equation number nineteen is substituted into the formula
number eighteen which is expressed as below (a, b).
gg(x)g(x) =
s2 [f(x2 ) − 2f(x1 ) + f(x0 )] s[−f(x2 ) + 4f(x1 ) − 3f(x0 )]
+
+ f(x0 ).
2
2
(25)
where, the remaining portion of equation number twenty illustrates that it is function
x third time estimation. Thus, expanding the both function × 2 f(x2 ) and × 1 f(x1 )
through Taylor’s series.
f(x2 ) = f(x0 ) + 2f (x0 )h + 2f (x0 )h2 + 0 h3 .
f(x1 ) = f(x0 ) + f (x0 )h +
f (x0 )h2
+ 0 h3 .
2
(26)
(27)
Through interchanging f(x2 ) and f(x1 ) in (25) with (26) and (27). The following
result is obtained.
g(x) = f(x0 ) + f (x0 )sh +
f (x0 )s2 h2
+ 0 h3 .
2
(28)
Hence h = x − x0 , is the Taylor’s series expansion for f(x) about x0 is
f(x) = f(x0 ) + f (x0 )sh +
f (x0 )s2 h2
+ 0 h3 .
2
Subtracting (28) from (29)
f(x) − g(x) = 0 h3
(29)
258
A. Q. Shahzad et al.
Thus, boundary terms and conditions specified through (24) outcomes in third
order approximation where x0 < x < x1 .
Same technique is utilized to get cN+1 . In case x is the interval xN−1 , xN , then
the boundary condition will be;
cN+1 = 3f(xN ) − 3f(xN−1 ) + f(xN−2 ).
(30)
Through the CC interpolation-kernel equation number fifteen along with the
boundary conditions equation number nineteen and formula number twenty-five.
Consequently, through this approach visualize complete framework of cubicconvolution interpolation-function which can now be given as [31].
When xk < x < xk+1 the cubic-convolution interpolation-function is
ck 3s3 − 5s2 + 2
ck−1 −s3 + 2s2 − s
+
g(x) =
2
23
3
ck+1 −3s + 4s2 + s
+ ck+2 s − s2 /2
+
2
where,s =
x−xk
h
and ck = f(kx ) for k = 0, 1, 2, . . . , N;
c−1 = 3f(x0 ) − 3f(x1 ) + f(x2 ) and cN+1 = 3f(xN ) − 3f(xN−1 ) + f(xN−2 ).
3.2.3
Kalabagh Seismicity
Kalabagh seismic zonation shows that it lies in moderate zone. As Fig. 5 shows 30%
lucidity whereas, Figs. 6 and 7 demonstrate zonation based on depth.
3.2.4
Kalabagh Cubic Convolution Interpretation
The study is focused on the earthquake distribution and hazard analysis of Kalabagh
area. The area hosts a major strike slip fault called the Kalabagh strike slip fault as
shown in Figs. 5, 6, and 7 respectively. The earthquake magnitudes are plotted and
then interpolated over this region by using Cubic Convolution techniques e.g., Cubic
Convolution on basis of different depth which ranges from (0–50) km, (50–100) km
and (100–250) km. The area has a maximum recoded range of earthquake magnitude
of greater than 3 Mw and less than 5.6 Mw which lies in the moderate earthquake
range in terms of magnitude as shown in Figures. Cubic Convolution gives us best
result and is therefore considered the best and advanced technique for zonation and
other image processing due to following reason.
UAV-Based Rescue System and Seismic Zonation …
259
Fig. 5 Map shows all events frequency over study area (Kalabagh Fault) with 30% transparency
over satellite imagery
Fig. 6 Map illustrates entire earthquake frequency up to 50 km depth
260
A. Q. Shahzad et al.
Fig. 7 Map illustrates entire earthquake frequency from 50 to 100 km depth
4 Conclusion
This study concludes that utilizing an Autonomous System based on Drone Rescue
(AUTO-SBDR) for post-seismic acute situation-management. Through this approach
many lives will be saved even evacuate from life-threatening condition and more
significantly decreasing response time to emergency. Additionally, model displayed
in this paper is secure, cheap, and consistent as matched to the existing methodology. Seismic-Zonation is utilized to identify earthquake prone areas. Furthermore,
zonation is conducted through kriging and cubic convolution which assist UAVs
to perform rescue operation. Seismological studies along with UAV-based rescue
system illustrate effective performance. Where, the proposed paper future work
comprises executing drones-based network along with sensors via LoRaWAN technology. Moreover, this approach utilizes smart-city and IoT paradigms. Furthermore,
this approach intends to adapt drones to perform other services independently in case
of critical conditions such as alarming about flood, earthquake precursor, detecting
fire and act as extinguisher.
References
1. M. Dilley, Natural Disaster Hotspots: A Global Risk Analysis, vol. 5. World Bank Publications
(2005)
2. I.K. Kouadio, S. Aljunid, T. Kamigaki, K. Hammad, H. Oshitani, Infectious diseases following
UAV-Based Rescue System and Seismic Zonation …
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
261
natural disasters: prevention and control measures. Exp. Rev. Anti-Infect. Ther. 10(1), 95–104
(2012)
B. Rabta, C. Wankmüller, G. Reiner, A drone fleet model for last-mile distribution in disaster
relief operations. Int. J. Disaster Risk Reduction 28, 107–112 (2018)
T. Zw˛egliński, The use of drones in disaster aerial needs reconnaissance and damage assessment–Three-dimensional modeling and orthophoto map study. Sustainability 12(15), 6080
(2020)
M. Hirschinger, No vehicle means no aid–a paradigm change for the humanitarian logistics
business model, in Essays on Supply Chain Management in Emerging Markets (Springer, 2016),
pp. 43–64
L.A. Haidari, S.T. Brown, M. Ferguson, E. Bancroft, M. Spiker, A. Wilcox, R. Ambikapathi,
V. Sampath, D.L. Connor, B.Y. Lee, The economic and operational value of using drones to
transport vaccines. Vaccine 34(34), 4062–4067 (2016)
J. Emblemsvåg, On probability in risk analysis of natural disasters. Disaster Prev. Manag. Int.
J. 17, 508–518 (2008)
D. Soesilo, P. Meier, A. Lessard-Fontaine, J. Du Plessis, C. Stuhlberger, V. Fabbroni, Drones
in humanitarian action. a guide to the use of airborne systems in humanitarian crises;
FSD Report (Swiss Foundation for Mine Action, Geneva, Switzerland, 2017), pp. 23–34.
Available online: https://irevolution.files.wordpress.com/2011/07/drones-in-humanitarian-act
ionemail.pdf (accessed on 28 May 2020)
Back to the Future: Drones in Humanitarian Action. Available online: https://blog.werobotics.
org/2019/11/12/back-to-the-future-drones-in-humanitarian-action/ (accessed on 1 Mar 2020)
R. Laing, M. Leon, L. Mahdjoubi, J. Scott, Integrating rapid 3D data collection techniques to
support BIM design decision making. Procedia Environ. Sci. 22, 120–130 (2014)
S. Verykokou, A. Doulamis, G. Athanasiou, C. Ioannidis, A. Amditis, UAV-based 3D modelling
of disaster scenes for urban search and rescue, in Proceedings of the 2016 IEEE International
Conference on Imaging Systems and Techniques (IST), Chania, Greece (4–6 Oct 2016), pp. 106–
111
O.S. Dharmaputra et al., The occurrence of insects, fungi and organoleptic characteristics in
stored coffee beans in Lampung. BIOTROPIA-The Southeast Asian J. Trop. Biol. 14 (1999)
J.W. McDougall, S.H. Khan, Strike-slip faulting in a foreland fold-thrust belt: The Kalabagh
Fault and Western Salt Range, Pakistan. Tectonics 9 (5), 1061–1075 (1990)
A.A. Khwaja, M. Qasim Jan, The 8 October 2005 Muzaffarabad earthquake: Preliminary
seismological investigations and probabilistic estimation of peak ground accelerations. Curr.
Sci. 1158–1166 (2008)
A. Hallam, J.B. Maynard, The iron ores and associated sediments of the Chichali formation
(Oxfordian to Valanginian) of the trans-indus salt range, Pakistan. J. Geol. Soc. 144(1), 107–114
(1987)
F. Ali, et al., Range front structural style: An example from Surghar range, North Pakistan. J.
Himalayan Earth Sci. 47 (2), 193 (2014)
A. Ali, Structural analysis of the trans-Indus ranges: Implications for the hydrocarbon potential of the NW Himalayas, Pakistan. Dissertation National Centre of Excellence in Geology
University of Peshawar, Pakistan (2010)
S. Obaid, A.W. Qureshi, I.A. Abbasi, Lithofacies, sand-bodies geometry and depositional
setting of the Datta formation in Surghar Range, North Pakistan, in SPE/PAPG Annual Technical
Conference. OnePetro (2005)
B. Wadood et al., Sequence stratigraphic framework of the Jurassic Samana Suk carbonate
formation, North Pakistan: Implications for reservoir potential. Arabian J. Sci. Eng. 46, 525–542
(2021)
F.A. Jassim, F.H. Altaany, Image Interpolation Using Kriging Technique for Spatial Data.
arXiv preprint arXiv:1302.1294 (2013)
S. Fadnavis, Image interpolation techniques in digital image processing: an overview. Int. J.
Eng. Res. Appl. 4(10), 70–73 (2014)
262
A. Q. Shahzad et al.
22. R.E. Rossi, J.L. Dungan, L.R. Beck, Kriging in the shadows: geostatistical interpolation for
remote sensing. Remote Sens. Environ. 49(1), 32–40 (1994)
23. M.R. Stytz, R.W. Parrott, Using kriging for 3d medical imaging. Comput. Med. Imaging Graph.
17(6), 421–442 (1993)
24. T. Hengl, G.B.M. Heuvelink, D.G. Rossiter, About regression-kriging: From equations to case
studies. Comput. Geosci. 33 (10), 1301–1315 (2007)
25. R. Keys, Cubic convolution interpolation for digital image processing. IEEE Trans. Acoust.
Speech Signal Process. 29(6), 1153–1160 (1981)
26. E. Meijering, M. Unser, A note on cubic convolution interpolation. IEEE Trans. Image Process.
12(4), 477–479 (2003)
27. S.E. Reichenbach, F. Geng, Two-dimensional cubic convolution. IEEE Trans. Image Process.
12(8), 857–865 (2003)
28. C. Lin et al., The efficient VLSI design of BI-CUBIC convolution interpolation for digital
image processing, in 2008 IEEE International Symposium on Circuits and Systems (IEEE,
2008)
29. T.-K. Truong et al., Image data compression using cubic convolution spline interpolation. IEEE
Trans. Image Process. 9 (11), 1988–1995 (2000)
30. S.K. Park, R.A. Schowengerdt, Image reconstruction by parametric cubic convolution. Comput.
Vision Graph. Image Process. 23(3), 258–272 (1983)
31. J. Shangguan et al., Fast algorithm of modified cubic convolution interpolation, in 2011 4th
International Congress on Image and Signal Processing, vol. 2 (IEEE, 2011)
Multi-sensor Fusion Methods
for Unmanned Aerial Vehicles to Detect
Environment Using Deep Learning
Techniques
Pradeep Duraisamy, Venkatesh Babu Sakthi Narayanan, Ramya Patturajan,
and Kumararaja Veerasamy
Abstract Sensor fusion is the ability to merge the inputs of multiple sensors like
radars and cameras to form a single image of the environment around a vehicle. The
output model brings an accurate image since it compares the strengths of the different
sensors. Deep learning provides some networks and techniques to avail sensor fusion
capability. Deep learning is the branch of Machine learning. Deep learning is based
on learning and improving on its own by examining computer algorithms. Sensor
fusion is possible through deep learning since sensors need to merge themselves to
get a clear and accurate output of the surrounding of a vehicle. In this study, I have
put together some techniques of deep learning based sensor fusion for detecting
environments around Unmanned Aerial vehicles.
Keywords Sensor fusion · Convolutional neural network · Multi-sensor fusion
algorithm · Fusion net · Deep learning
1 Introduction
Autonomous Vehicles have a high higher level of potential to reduce the risk of
dangerous driver behavior. Automation helps to reduce the count of accidents through
crashes on our roads. Recently, Convolutional Neural Networks are mainly proposed
to establish the accuracy of the camera images. The fusion of sensors using Neural
Networks is encompassed in order to get more accurate object detection. Let us see
the detailed techniques of Deep Learning [1] that club together various sensors.
P. Duraisamy (B)
M.Kumarasamy College of Engineering, Karur, Tamil Nadu 639113, India
V. B. S. Narayanan · R. Patturajan
Christian College of Engineering and Technology, Oddanchatram, Tamil Nadu, India
K. Veerasamy
K.Ramakrishnan College of Engineering, Trichy, Tamil Nadu 621112, India
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_15
263
264
P. Duraisamy et al.
1.1 Deep Learning in Object Detection
The intertwined outputs of various sensors are accepted to obtain a better outcome
than single-sensor inputs. To go along with this, the principle combination procedures
featuring benefits and disadvantages are scrutinized. Inhabitance Grids: Probabilistic
inhabitances lattices (POGs) are adroitly the least complex way to deal with execute
Bayesian information combination techniques. Albeit basic, POGs can be concerned
with various issues inside the insight task: e.g., planning, traversing article location,
and sensor combination.
KF (Kalman Filter): Kalman Filter highlights build it fit for managing multiple
sensor assessment also information combination issues. In the first place, its unequivocal depiction of cycles and perceptions permits broad sensor facsimiles that are fused
inside the essential calculation. Next, the predictable utilization of factual proportions of vulnerability constructs conceivable to computable assesses the job every
sensor take part in that and large framework execution.
MC Methods (Monte Carlo): Monte Carlo strategy is appropriate to deal with
the issues from state change methods and perception methods are profoundly nonstraight. Its justification is the example related strategies could address exceptionally
broad likelihood densities. Specifically, multi-modular or various theory thickness
capacities are very much taken care of by Monte Carlo procedures.
IC technique (Interval Calculus): In the IC technique, vulnerability is addressed
by leap qualities. The significant benefit contrasted with probabilistic technique is
that IC gives better proportions of vulnerabilities without any likelihood data yet the
blunders of sensor information are limited to a specific worth. In any case, IC is not
for the most part utilized in information combination in light of the trouble to get
results that merge to an ideal worth; and the trouble to encode conditions between
factors that are at the centre of numerous information combination issues.
Logic on Fuzzy: This logic rationale is a well-known technique in charge and
detector combination to address vulnerability point thinking depends on levels of
truth as opposed to outright esteem. However, this strategy turns out to be more
mind-boggling with the expansion of detector results. Additionally, approval of this
strategy requires broad examination that wellbeing is a significant thing.
Proof Theory (ET): The upside of ET is its capacity to address fragmented proof,
absolute obliviousness together with the absence of requirement as deduced possibilities. The field of astute automobile discernment is an assortment of blemished
data: unsure or uncertain. By its nature, objects are lost (impediments), the detector
can’t gauge every important quality of the item (equipment constraints), and when
perception is questionable (incomplete article discovery). In any case, with a larger
number of speculations ET turns out to be less computational manageable.
Multi-sensor Fusion Methods for Unmanned Aerial Vehicles …
265
2 Multiple Sensors Fusion with CNN
It is seen that the precision in distinguishing and limiting a moving item actually
needs heartiness when natural and environmental factors change [2, 3]. Multi-sensor
information combination is the most common way of joining a few perceptions from
various sensor contributions to give a more complete, vigorous and exact portrayal
of the climate of interest. Figure 1 illustrates the process of a Convolutional Neural
Network that incorporates the data from multiple cameras to obtain a final instruction
on how to drive the car by regulating the surrounding environment.
In this image, the radar together with ultrasonic detector information are used
to distinguish the interval as the article that is to be taken care of the principal
extent of detector combination in order to work on the vigour about identification
as broad scope as interval together with detector skyline. Camcorder pictures exist
changed over into dot format information that is prepared to eliminate the frontal
area. The picture information is fragmented together with afterwards totally divided
information are organized as a variable quantity. The tensor information is then
melded with the interval recognition information. The composite variable quantity
is next additionally handled to streamline the jumping box. Now, the information is
prepared to utilize the profound studying calculation through CNN.
Fig. 1 Convolutional neural network for autonomous vehicle
266
P. Duraisamy et al.
2.1 ADAS (Advanced Driver Assistance System)
Convolutional Neural Networks (CNNs) seems effective for profound learning calculations for characterizing two and three-dimensional pictures, be that as it may, the
implementation in Advanced Driver Assistance System and independent automobile
is as yet is at outset. The latest job that is fairly stable is an exact assessment of
profound studying on genuine driving information in constant for both errands: path
recognition and vehicle location. ADAS is that Convolutional Neural Network can
give a satisfactory answer as the previous undertakings by running at outline rate
that is needed for an ongoing frame framework. In another new work, creators utilize
profound learning for recognizing people on foot, not withstanding, creators’ remarks
that albeit the presence of profound organization is superior to course calculations
for identifying complex examples, it is delayed for an ongoing frame walker location [4]. No current work concentrates on sensor combination alongside profound
nonpartisan organization, which is the focal point of ADAS.
General guideline as a conventional Convolutional Neural Network calculation
has beneath. Convolutional Neural Network is a multi-facet perceptron that switches
back and forth in the middle of the convolutional and pooling surfaces for each
information channel. It tends to be constrained by changing their profundities and
broadness, contrasted and comparative estimated layers CNNs have fewer associations and boundaries to prepare. Convolutional layers comprise little learnable channels that convolve or slide over the information picture. In the forward cycles, we
slide each channel across the width and tallness of the info picture and work out the
speck item at a position. The Pooling layer helps us dimensionality decrease of the
information picture which helps in computational intricacy and furthermore helps
in diminishing the over fitting. It additionally has a completely associated layer that
interfaces with the whole info volume as in a common Neural Network. Figure 2
outlines an illustration of CNN based item identification.
In ADAS, CNN has been used for the discovery of bicycles, vehicles, passersby,
and streets by utilizing Google TensorFlow, a famous open-source library for
profound learning. The library upholds numerous CPUs or GPUs alongside various
programming dialects like Python, C++ and so on it contains the execution of CNN
calculation, which is profoundly nonexclusive; the learning model is addressed as
information stream diagrams where hubs are called activities and every activity
accepts input as tensor and yields as a tensor. A client can set boundaries for network
design, and furthermore boundaries like learning rate, group size and a number of
cycles and so on for adjusting under-fitting and over-fitting of the classifier.
One more key component in TensorFlow is its help for pre-preparing through
Inception, Google’s open-source picture classifier which is prepared on 1.2 million
pictures with 1000 distinctive class marks in more than 2 weeks with 8 GPUs. Prepreparing works like an exchange learning worldview, with the end goal that the
examples gained from Inception information, can be moved to an application that
has its own arrangement of preparing information occurrences. The retraining system
Multi-sensor Fusion Methods for Unmanned Aerial Vehicles …
267
Fig. 2 Object identification using convolutional neural network
uses existing boundaries that have been learnt as a feature of Inception characterization and henceforth utilizes this recovery significant time in preparing. Plus, it
helps as far as preparing information need; a profound exactness classifier can be
worked with less preparing information by taking advantage of the exchange learning
worldview.
3 Multi-sensor Fusion Algorithm
The proposed algorithm of Multi-sensor fusion confronts object tracking. It classifies
the difficulties to traverse at different speeds and designs ways to implement learning
in the process of detection. The steps in the algorithm are as follows.
Step 1
Detecting Moving Objects
Unlabeled data are sorted from the camera captured earlier. The distance of an
indicated object is determined and error is calculated found from the sensor data.
The size of the object is estimated by perception.
Step 2
Data Fusion from sensors
The blend of sensor information acquired at the location level can be utilized
to decide the various properties of an item. Distinctive probabilistic strategies are
adjusted to test the information acquired from sensors to track down a composite
portrayal of the best theories in a sliding window of time. The combination technique
is dynamic to deal with the transient changes with the item depiction.
268
P. Duraisamy et al.
Step 3
Learn information consistently followed
The profound learning technique includes taking in information from various
degrees of data gathered from the past yield. It utilizes complex machine vision
ways to deal with training the information from various course layers.
3.1 Sensor Fusion Using FusionNet
Fusion Net architecture [5] combines the feature maps of different sensors to detect
the object. The network design has the feature extractor which comprises feature
maps followed by detection heads and the maps are of multiple scales. The method
observes multiple sensors like cameras, radar and liadars and compresses all its
inputs to produce accurate output. The objective of FusionNet is to concentrate
and consolidate highlights removed from various sensors noticing a similar space,
from a possibly unique perspective, and with their general positions known. Each
component extraction branch fuses a spatial change to such an extent that the yield
highlight maps from each branch are spatially lined up with different branches.
3.1.1
Fusion Net Architecture
FusionNet consists of two branches, the Radar branch, which measures the reach
azimuth picture from the radar, and the Camera branch that measures the pictures
caught by a front oriented camera. After the free component extractor branches,
these elements are then gone through the combination layer(s) to guarantee that the
organization learns.
Radar
Fusion Net isn’t a point cloud. The radar branch [6] takes a thick 2D territory
azimuth “picture”, permitting to utilize highlight pyramid network structures famous
in picture object location organizations. Since the objective is to anticipate bouncing
boxes in Cartesian facilitates, a planning layer is added to the middle component
maps. The fixture of spatial change right off the bat in the middle of the road include
layers gave the best exhibition. After this change, more convolutional layers were
added before connection with different branches.
Camera
To change the camera picture into the Cartesian space, an Opposite Projection
Mapping [8] is conceived, which is a homography change of the camera picture.
To process this projection planning, in the beginning, the camera is imaging a planar
scene (for example the radar plane, which is around corresponding to the street plane).
Then, the characteristic and outward alignment data to extend a bunch of focuses in
the Cartesian radar plane to picture organizes is used. A planar homography change
Multi-sensor Fusion Methods for Unmanned Aerial Vehicles …
269
would then be able to be found by utilizing the standard 4-point calculation. In
the occasion that alignment isn’t accessible, it is additionally conceivable to physically appoint various tie focuses, at last settling for the best homography utilizing a
least-squares strategy.
The design of the camera branch is basically the same as the radar branch. Be
that as it may, rather than performing the organized change in the component maps,
the organization played out the best when this change is applied straightforwardly
to the camera picture rather than the highlight maps. After the homography change,
the information picture to the organization is a 3-channel 256 × 256 shading picture.
The picture directions ought to now match the Cartesian radar picture organizes in
the event that the planar supposition is right and the camera doesn’t move as for the
radar.
Data Fusion
For object location, we applied SSD [7] heads on the combination highlight maps.
Anchor boxes were picked to coordinate with the appropriation of the ground truth
boxes from our preparation set. K-implies bunching was utilized to develop a bunch
of anchor boxes that are more reasonable for our vehicle identification organization.
It ought to be evident that the essentially zeroing in on vehicles, there are just a
modest bunch classes of vehicles that are ordinarily out and about (e.g., fair size car,
truck). Specifically, there is a low minor departure from the width of these vehicles
given the limits that are forced by the US Department of Transportation on the path
widths.
3.1.2
Data Preprocessing
The radar sensor yields an inadequate 2D point cloud with related radar characteristics. The qualities of the radar return are put away as pixel esteems in the expanded
image. The input camera picture comprises three channels (red, green, blue); to this,
we add the previously mentioned radar channels as the contribution for the neural
network. We link the point mists of the three sensors into one and utilize this as the
projected radar input source.
The more prominent assortment in the thickness of the radar and the camera information—in contrast with the lidar and the camera—represents the test of tracking
down an appropriate method to intertwine the information in one shared organization
structure. To manage the sparsity of radar information, employes probabilistic matrix
guides to create ceaseless data from the radar.
First and foremost, we don’t reimburse the development of various things when we
connect the past radar situated in the data. As the Scenes dataset is named at 2 Hz, no
ground-truth is open for moderate radar area cycles, radar object acknowledgments
simply present in center cycles are maybe filtered through. Second, slight spatial
miscalibrations between the radar and camera sensors achieve misalignment of the
radar revelation regions and the ground-truth hopping boxes at more imperative
distances. Third, the data from the radar and the camera are not recorded at definitively
270
P. Duraisamy et al.
Fig. 3 Multi-sensor fusion network
a similar time. Fourth, while the radar distance assessment is genuinely reliable, its
assessments are not magnificent and slight mistakes can make the acknowledgments
lie outside of the ground-truth bouncing boxes.
3.1.3
Architecture of Network Fusion
The neural organization engineering expands on RetinaNet with a VGG spine [8].
The organization is reached out to manage the extra radar channels of the expanded
picture. The yield of the organization is a 2D relapse of bouncing box facilitates and
a characterization score for the bounding box. The organization is prepared to utilize
central misfortune. The pattern technique utilizes a VGG include extractor during
the first convolutional layers.
Figure 3 illustrates the combining method of camera and lidar projections into the
segmented object output using Convolutional Neural Network. The images identified
in the first convolutional network are consequently sent to the second convolutional
layer for the purpose of segmentation. The accuracy of the object image is very high
when the second layer segments and projects the output.
3.2 Sensor Information Fusion Technology
Multi-rate Sensor Information Combination Strategy (MRSIFS) depends on multidimensional convolution square [9] and time-repeat examination development, which
Multi-sensor Fusion Methods for Unmanned Aerial Vehicles …
271
execute multichannel equivalent inadequacy feature extraction, and the components from unrefined signs with different testing rates are used for issue investigation. The multi-rate sensor performs different undertakings feature extraction units
using a multidimensional convolution square and Adam adversity work, which basically further develops the part extraction limit. Finally, the proliferation stage’s test
outcomes show that the proposed perform different errands model achieves higher
examination precision than the current strategies. Plus, the manual component determination for each assignment is superfluous in MRSIFS, which has the potential
toward a universally useful system.
In the multi-rate sensor feature extraction stage, the deficiency feature sepaevaluated by the data layer is dealt with into the multidimensional convolution feature
learning block. The segments of the convolution layer have different estimations. As
we presumably know, convolution parts of different mea-surements go about as channels of different objective scales to remove weakness features in the unrefined signs
and simultaneously eliminate the arrangements of information signals in different
repeat gatherings.
Arrangements of the convolution part estimation are joined through the association
layer to shape a multi-rate incorporate guide. As an issue information authority, the
connection layer can add up to arrangements of different scales to shape a multi-rate
feature set. It very well may be seen that the data from various sensors is changed
into various channels through the information layer, and afterward, by then, the issue
features are gained through the component extraction layer and component mix layer
progressively. Specifically, the information layer of the proposed methodology is a
three dimensional system, and the normal outcomes are taken as the yield of the
perform different assignments model.
3.3 CNN and Regression
The 2D CNN is made of six segments, including STFT, input layer, downsampling
layer, smooth layer, and upsampling layer. The downsampling layer is made out of K
expectation blocks, and every one of them contains three convolutional layers. The
upsampling layer is completed ward on the bilinear. Additionally, the smooth layer
contains a lone convolutional layer. To eliminate the concealed issue information in
the sign past what many would consider conceivable, four downsampling layers, four
smooth layers, and three upsampling layers were dynamically used in the preliminary
model with the dataset. The arrangements are dealt with into the yield layer and fix
the yield of the last layer as the issue incorporate isolated by the 2D convolution part.
It ought to be seen that in the 2D convolution section, a significant convolutional network structure is utilized. Along these lines, though significant level issue
arrangements can be taken out, consenting to some current examinations [10], the
significant level convolutional association may lose some shallow level inadequacy
features. In this paper, the downsampling layer, smooth layer, and upsampling layer
are used to facilitate, with the objective that the components removed from the
272
P. Duraisamy et al.
shallow convolutional association can be joined into the subsequent significant level
part. Information part fix the yield of the last layer as the weakness feature extri-cated
by the 2D convolution.
Differentiated and the two-dimensional convolution structure, the onedimensional convolution structure simply holds the piece of the lower looking at
layer, and a couple of changes are made in the convolutional network limit settings.
The DNN net-work block is made out of four full affiliation layers. Basically, the
backslide layer includes three full affiliation layers, with ReLU as an activation work.
Limit nuances and plan execution of DNN and backslide layer are shown in independently. After each convolutional layer in the data layer and downsampling layer,
and the totally related layer in DNN, bunch normalization is utilized to accelerate
the arrangement pattern of MRSIFS.
4 Conclusion
With the huge progression of sensor and correspondence innovation and the solid
utilization of deterrent location procedures and calculations, robotized driving is
turning into an urgent innovation that can change the eventual fate of versatility.
Sensors are principal to the impression of vehicle environmental elements in a
computerized driving framework, and the utilization, what’s more, execution of
various coordinated sensors can straightforwardly decide the wellbeing and attainability of computerized driving vehicles. Hypothetical logical writing on information
combination began to show up with executions and more algorithmic advancements.
Deep learning based sensor fusion techniques are discussed in this paper.
References
1. Felix, A Deep Learning-based Radar and Camera Sensor Fusion Architecture for Object
Detection. arXiv:2005.07431v1 [cs.CV] 15 May 2020
2. Caio, Environment-aware sensor fusion using deep learning, in ICINCO 2019—16th International Conference on Informatics in Control, Automation and Robotics
3. Yibing, Environmental Perception and Sensor Data Fusion for Unmanned Ground Vehicle.
https://doi.org/10.1155/2013/903951
4. Zhangjie, Infrared–ultrasonic sensor fusion for support vector machine–based fall detection. J.
Intell Mater Syst Struct 29(9), 2027–2039 (2018)
5. D.J. Yeong, Sensor and sensor fusion technology in autonomous vehicles: A review. Sensors
21, 2140 (2021). https://doi.org/10.3390/s21062140
6. T.-Y. Lim, Radar and camera early fusion for vehicle detection in advanced driver assistance
systems, in Machine Learning for Autonomous Driving Workshop at the 33rd Conference on
Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada
7. Behnaz, A decision fusion method based on multiple support vector machine system for fusion
of hyperspectral and LIDAR data. Int. J. Image Data Fusion 5(3), 196–209 (2014). https://doi.
org/10.1080/19479832.2014.919964
Multi-sensor Fusion Methods for Unmanned Aerial Vehicles …
273
8. R. Sekhar, Object detection from a vehicle using deep learning network and future integration
with multi-sensor fusion algorithm. SAE Technical Paper. https://doi.org/10.4271/2017-010117
9. Xiangkai, A multirate sensor information fusion strategy for multitask fault diagnosis based
on convolutional neural network. J. Sens. (9952450) (2021). https://doi.org/10.1155/2021/995
2450
10. H. Pan, Deep learning based data fusion for sensor fault diagnosis and tolerance in autonomous
vehicles. Pan et al. Chin. J. Mech. Eng. 34, 72 (2021). https://doi.org/10.1186/s10033-021-005
68-1
General Parametric of Two
Micro-Concentrator Photovoltaic
Systems for Drone Application
Sarah El Himer , Mariyam Ouaissa , Mariya Ouaissa ,
and Zakaria Boulouard
Abstract Solar powered high altitude long endurance drones can be a solution for
many missions. The difficulty in making such drones is due to the very high altitudes
targeted and the low electric power available to power the motors. The idea in this
paper is to have powered these motors by concentrated solar energy to waste space
and high power. This article presents the details of the optimization of the primary
and two imaging secondary optic which are circular and square dome used in the
micro CPV system, this system which will be mounted on a real drone. The optical
performance of the two designed micro-CPV systems is assessed by ray tracing
simulation. Subsequently, the performance of the hybrid micro CPV systems based
on the lens associated with the domes is tested. As results; we found that the high
optical efficiency, the large acceptance angle and the power achieved by the Micro
CPV system using the square dome (88.72%, 1.4° and 140 W). Concerning the flux
distribution we got that the flux distribution given by the square dome is uniform.
Finally, the power requirement to power the drone’s motor has been roughly estimated
at 390 W for a period of around half an hour. To this must be added the supply of
power to the on-board electronics, the needs of which should not exceed a few watts.
Our need to satisfy this power is 3 units composed of a Fresnel lens and a square
dome.
Keywords Micro CPV · Dome · Optical efficiency · Acceptance angle
S. E. Himer (B)
Sidi Mohammed Ben Abdellah, Fez, Morocco
e-mail: sarah.elhimer@usmba.ac.ma
M. Ouaissa · M. Ouaissa
Moulay Ismail University, Meknes, Morocco
e-mail: mariyam.ouaissa@edu.umi.ac.ma
Z. Boulouard
LIM, Hassan II University, Casablanca, Morocco
e-mail: zakaria.boulouard@fstm.ac.ma
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
M. Ouaissa et al. (eds.), Computational Intelligence for Unmanned Aerial Vehicles
Communication Networks, Studies in Computational Intelligence 1033,
https://doi.org/10.1007/978-3-030-97113-7_16
275
276
S. E. Himer et al.
1 Introduction
In recent years, drones have experienced a growing craze in the field of civil and
military aeronautics. We can differentiate the many drones according to their size
(from 77 mm to 77 m wingspan), according to their autonomy (from a few minutes
to several days) or even according to their mode of movement (forward flight or
hovering flight) or finally according to their number of propellers (from 1 to 4). The
miniaturization of a drone brings multiple constraints on each of its components: on
aerodynamics (operation at low Reynolds numbers, a field not well known), on its
propulsion (small but powerful engines, with good efficiency), on its flight dynamics
(more sensitive to atmospheric disturbances) and its on-board payload (light and
functional).
Drones or Unmanned Aerial Vehicles (UAVs) are unmanned flying vehicles
capable of carry out a mission more or less autonomously. Their main use is military
for reconnaissance or surveillance missions, without risk of loss of life. Indeed, they
are well suited for carrying out missions which would potentially endanger a crew
or which require a permanent presence in the area which would be tedious for a crew
on board. Their use began with everything related to observation and then extended
to the acquisition (or even illumination) of targets as well as electronic warfare, and
the destruction of targets. From now on, civilian applications are emerging such as
highway traffic monitoring, forest fire prevention, meteorological data collection or
even the inspection of works of art.
Their size varies from centimeters to several meters, as does their mode of propulsion, which changes according to needs (electric motors for miniature drones, thermal
or jet engine for larger drones). There are two types of vectors: fixed wings for forward
flight and rotary wings for hovering flight.
The history of UAVs begins in 1883 when Douglas Archibald attached an
anemometer to a kite. He succeeded in measuring wind speed at altitudes of 400 m
[1]. Five years later, Arthur Batut fitted a kite with a camera and made the first aerial
photo on June 20, 1888 in Paris [2]. These were the first flying machines equipped
for surveillance or detection.
Two unmanned flying vehicles were in development at the end of the First World
War: “the flying bomb” of the Navy [1] and the “Kettering Bug” of Charles Kettering
[2].
In 1935, Reginald Denny, former of the British Air Force and expatriate in the
United States, conceived the first model which was going to know the war: the RP-1.
The intensive use of reconnaissance drones began with the Vietnam War. From
1965 to 1972, drones monitored areas in China, North and East Vietnam, where it
was too dangerous to send aircraft with pilots. During the Gulf War, six Pioneers
participated and collected real-time data on the reconnaissance and surveillance of
Iraqi troops. Other functions of drones were exploited as that of drone-bait. During
the first night of American attack, the drones “BQM-74C Chukar” were used to
disrupt the anti-aircraft defense by creating, by their fuselage, a radar signature close
to that of the B-52 bombers [3].
General Parametric of Two Micro-Concentrator …
277
The success of the use of drones during this war prompted DARPA to launch
several programs to finance the development of other drones: tactical drones for
operations on the front line, medium-range drones to monitor the area of operations,
long-range and long-range drones to search for more distant targets and discreet
drones to observe very well-guarded areas [4]. With widening the application scope
of unmanned aerial vehicle (UAV) as the driving force, the development of solarpowered UAV recently has attracted more attention in academia and commercial
industries. A critical factor limiting the scope of application of conventional batterypowered electric UAVs is their energy storage capability. Among the solar technologies suitable for this application are concentration photovoltaic (CPV) and Micro
CPV.
Concentrated Photovoltaic (CPV) is one of the most promising solar-energy-based
methods for generating sustainable electricity. It is a high-efficiency replacement
for the conventional flat plate module [5]. This method provides for an important
reduction in the amount of space required in the cells, the recent halt of activity of
several CPV players has posed a challenge to producing a given amount of electrical
power while lowering its cost through high efficiency and moderate expenses. This
commercial failure contrasts with the technological success that has been attained in
the field, with modules nearing 40% efficiency [6].
The goal of employing micro-CPV is to lower cell size to the sub-mm range
in order to increase a variety of benefits while also presenting certain production
problems due to the small cell size. The micro-CPV shrinks CPV modules since
their thickness, weight, and volume are related to the cell size; hence, the module’s
width is reduced.
One or two optical components can be used to make concentration optics. Several
writers have utilized various Fresnel lenses [7–22], parabolic mirrors [23], and
compound parabolic concentrators [24, 25]. The results demonstrate that using a
single stage for CPV systems can’t meet the general performance criteria; either
the system is extremely efficient but the flux distribution on the cell isn’t homogenous, or the converse is true. Then multistage concentrators, which are concentration
systems with two optical components, the first of which is known as the Primary
Optic (PO) and the second as the Secondary Optic (SO), can be more efficient.
According to the literature, For a 1000× desired concentration ratio, we discovered various systems based on the Fresnel lens, with secondary systems such as
the Pyramid, the Crossed Compound Parabolic Concentrator, Dome and Hyperbole,
which were explored by Rodrguez et al. [26, 27]. The four concentrators have an
acceptance angle of less than 1° and an optical efficiency of more than 80%, according
to simulation findings. With the pyramid, the best flux uniformity may be achieved.
Experiments confirmed these findings, with the optical efficiency falling to 73% and
the acceptance angle falling to 0.8°. El Himer et al. [28] evaluated and assessed four
CPV systems based on four secondary optical components (CPC, Cone, CCPC, and
pyramid) connected with a circular flat Fresnel lens for a 1000× as concentration
ratio using just simulation. They also demonstrated that, regardless of the material
employed, the pyramid as SO provides the most uniform lighting on the solar cell,
the widest acceptance angle (1.4°), and the maximum optical efficiency (83%). The
278
S. E. Himer et al.
performance of parabolic mirrors as a main or SO has been studied by several writers
[29, 30]. The results demonstrate that using a parabolic mirror as a major optical
element produces intriguing optical results, especially when combined with a flat
mirror, as it achieves the highest optical efficiency of 85 percent. When two optical
components of concentration are used, the CPV system is more efficient because it
allows for high optical efficiency, high concentration, and a broad acceptance angle.
Multiple systems have been investigated in the MCPV system for several years
[31, 32], and recent advancements have led to improved micro-concentrator ideas
and modules [33, 34]. Ritou [33] investigated a system consisting of a 19 × 19 mm2
conventional Plano convex lenses as the main optic (PO) and a SO, a dome lens with
0.6 mm2 cells as the SO and a concentration factor of 1000×. Its findings revealed
that the acceptance angle was around 0.7° [34]. With a Plano convex lens POE of
25 mm and a cell aperture of 0.25 mm, Jared MCPV design in [35] reaches 100
with an acceptance angle of 0.57°. The diameter of each square lens in the font lens
array in the Partric research is 2.0 mm, and each multi-junction cell is 0.0632 ×
0.0632 mm, giving the system a geometric concentration of 0.0632 mm of 1000×
. These results show that the system efficiency is about 92% while the acceptance
angle is 0.8° [36]. Salima et al. [37] offered a comparison of two MCPV systems:
the first uses a Fresnel lens as the main optic (PO) and a dome lens as the SO. The
second is made up of two lenses. The primary is Plano-convex, while the secondary
is formed like a dome. Polymethyl methacrylate (PMMA) and polycarbonate are two
potential materials (PC). The results reveal that the system consisting of a PMMA
Plano convex lens and a dome has the best optical efficiency in terms of flux nonuniformity; nevertheless, if the two systems are built of Polycarbonate, they allow
for a broad acceptance angle (PC).
This article presents the details of the optimization of the primary and two imaging
SO which are circular and square dome used in the micro CPV system, this system
which will be mounted on a real drone. The optical performance of the two designed
micro-CPV systems is assessed by ray tracing simulation. Subsequently, the performance of the hybrid micro CPV systems based on the lens associated with the domes
are tested.
After this introduction, Sect. 2 describes the designed prototype of solar optic
used; Sect. 3 reports the simulation results of our system, proposed in this study and
Sect. 4 presents the conclusions.
2 File Basic Concepts for Solar 4 Concentrators
Figure 1 depicts a typical optical system that separates two mediums with refractive
indices of n1 and n2 for the entrance and exit environments, respectively.
The system’s overall concentration ratio is defined as:
C = η × Cg
(1)
General Parametric of Two Micro-Concentrator …
279
Fig. 1 General view of an optical system
where η denotes the optical efficiency given by:
η=
Pout
Pin
(2)
and Cg the geometrical concentration is determined by the ratio of the entry and
exit surfaces, which is given by:
Cg =
Ain
n 2 sin θout
=
Aout
n 1 sin θin
(3)
3 Simulation Results
To effectively choose the elements of a microdrone, it is necessary to know the type
of mission for which it is made and the environment in which it must operate. In all
cases, it is intended for local missions. We will favor fixed wings for surveillance in
the area, and rotary wings for reconnaissance in difficult environments (city, forest)
where low movement speeds are required. In addition, it is better to use micro drones
of large size (50 to 70 cm) outdoors rather than small size (5 cm), more sensitive to
climatic conditions.
Finally, the need for acoustic discretion (or not) will guide the final choice of the
type of propulsion: a surveillance or reconnaissance mission needs to be silent, while
engineering structure inspection missions do not require this constraint.
Currently, for the new military needs, the microdrones will have to equip the
infantrymen as an aid in their progressions in hostile environment. Their main use
is of the type: “to see behind the hill or the building”. The expectations for these
vehicles are as follows:
280
S. E. Himer et al.
• Capable of moving easily by combining speed and relative insensitivity to bad
weather,
• Capable of stabilization, or even stop for shooting and other information, with
analysis and transmission,
• Robust, easily transportable and risk-free, simple and intuitive to use.
The drone (see Fig. 2) chooses with a size of 5 cm, as well as the circle that we
will put our optical system is 2 cm.
The Fresnel lens is made of a series of narrow concentric grooves. In this part, we
will consider the case of a two-dimensional Fresnel lens with grooves inward. Based
on Fig. 3, we can draw the following equations.
β = r + θi
(4)
d
2f
(5)
θi = ar tg
Fig. 2 Drone en bois
Fig. 3 Prism of the Fresnel
lens oriented inward
General Parametric of Two Micro-Concentrator …
281
Fig. 4 Designed fresnel lens
d
tan r = d 2
2 n 2 + f2 −1
(6)
The above equations allow us to design the 3D models of the Fresnel lens with
inward grooves shown in Fig. 4 where we have taken as an example a Fresnel lens
of 20 mm in diameter, a focal length of 20 mm and of 20 grooves.
In this section we present the dome with square and circular exit as an imaging
SO, The objective behind this SO (see Fig. 5) is to construct its entry surface in
such a manner that every ray coming from the main lens A(x1 ,y1 ) hits the cell B’s
opposite extreme point (x2 ,y2 ). Furthermore, all rays that travel through point A and
B, regardless of their angle of incidence, must follow Fermat’s principle (Eq. 15) of
optical path length conservation [33]. In these circumstances, Fermat’s premise is as
follows:
l1 + n S O E l21 = cst
l1 =
(x − x1 )2 + (y − y1 )2
(7)
(8)
282
S. E. Himer et al.
Fig. 5 Fermat principle applied to the design of secondary
l2 =
(x − x2 )2 + (y − y2 )2
(9)
The beginning point for constructing the profile is the dome height hSO . It is fixed
by forcing the bidimensional profile’s tangent to be horizontal at the origin, where
F/# is the main lens’s f-number.
(x0 , y0 ) = (0, h S O ) =
x0
sin ar ctg 2F/1 =
tan ar csin
nSO
(10)
Table 1 presents the optical efficiency, acceptance angle and the power en W
using circular and square dome, we observe that the high optical efficiency, the
large acceptance angle and the power achieved by the Micro CPV system using the
square dome (88.72%, 1.4° and 140 W). Concerning the flux distribution illustrated
in Fig. 6, we observe that the flux distribution given by the square dome is uniform
this is because the output shape of the square dome and the square shape of the cell
conforms on the other hand the circular shape of the dome cannot have a uniform
flow distribution. According to our research for several shapes of optical elements,
Table 1 Optical efficiency
and acceptance angle of
micro CPV module with
dome as SOs
Circular dome
Square dome
Optical efficiency (%)
85.69
88.72
Acceptance angle (°)
1
1.2
Power en W
105 W
140 W
General Parametric of Two Micro-Concentrator …
283
Fig. 6 Flux distribution of the dome
we always find that the square shape is always the right choice for the information
of the flux.
Next, we analyzed the position of the SO from the Fresnel lens. The main idea is
to find the optimal position of the SOs regarding the Fresnel lens.
Following geometrical considerations, three possible positions are a priori
possible without any optical losses: The first one, z = f, is defined by the focal
point of the lens (Fig. 7a), the second and the third are z = zmax and z = zmin , respectively defined by the place where the incoming flux size fit with the input diameter
(or side) of each element beyond and before the focal point of the lens, as shown on
Fig. 7b and c. These positions are defined by:
Z max = f +
A
tan θi
(11)
Z min = f −
A
tan θi
(12)
The changes in optical efficiency of the two systems Vs the relative location of the
SOs to the lens are shown in Fig. 8. It is apparent that the optical efficiency of square
and circular domes begins to decrease and then becomes constant over 1.5 mm in
the case of square dome and 1 mm in the case of the circular dome, then increases
as the distance between them increases when approaching to the maximal distance
from the lens z = zmax . The square dome keeps the high optical efficiency.
The acceptance angle of a concentrator is defined as the angle at which the optical
efficiency decreases to 80% of its original value [8]. For this, we ran simulations
284
S. E. Himer et al.
(a)
SO at the focal point, z=f
(b)
SO at the position maximal, z =zmax
(c) SO puted at the minimal position, z =Z min
Fig. 7 Position of secondary optic
with incidence angles ranging from 0° (normal incidence) to 1.4° (actual most solar
tracker precision).
Figure 9 illustrates the changes in the acceptance angle of the complete systems
(FL + SO) as the efficiency drops to 80% of its starting value. These findings demonstrate that the exaggeration has the greatest angle (1.2°) for a diameter of 2 mm for
square dome, concerning the circular dome; we obtain just 1° for all positions.
Figures 10 and 11 show the flux distribution on TJ solar cell using the two SOs, we
observe that the square dome keep its uniformity for all positions but using circular
one, we observe a pic with high intensity in the center of TJ solar cell, the circular
dome, the high intensity showed in minimal distance.
The power requirement to power the drone’s motor has been roughly estimated at
390 W for a period of around half an hour. To this must be added the supply of power
to the on-board electronics, the needs of which should not exceed a few watts. Our
need to satisfy this power is 3 units composed of a Fresnel lens and a square dome.
General Parametric of Two Micro-Concentrator …
Fig. 8 Optical efficiency versus position of secondary optic
Fig. 9 Acceptance angle versus position of secondary optic
285
286
S. E. Himer et al.
Fig. 10 Flux distribution using the circular dome as SO versus position of secondary optic
Fig. 11 Flux distribution using the square dome as SO versus position of secondary optic
4 Conclusion
In this study we have presented the details of the optimization of the primary and two
imaging SO which are circular and square dome used in the micro CPV system, this
system which will be mounted on a real drone. The optical performance of the two
General Parametric of Two Micro-Concentrator …
287
designed micro-CPV systems is assessed by ray tracing simulation. Subsequently,
the performance of the hybrid micro CPV systems based on the lens associated with
the domes is tested. We have obtained as results that the high optical efficiency, the
large acceptance angle and the power achieved by the Micro CPV system using the
square dome (88.72%, 1.4° and 140 W). Concerning the flux distribution we got
that the flux distribution given by the square dome is uniform. Finally, the power
requirement to power the drone’s motor has been roughly estimated at 390 W for
a period of around half an hour. To this must be added the supply of power to the
on-board electronics, the needs of which should not exceed a few watts. Our need to
satisfy this power is 3 units composed of a Fresnel lens and a square dome.
References
1. J.M. Bacon (1846–1904), The dominion of the air: the story of aerial navigation. The World
Wide School, Nov 1997
2. Batut, Photographie aérienne par cerf-volant. La Nature, vol. 1897, Vingt-cinquième année,
premier semestre: n° 1227 à 1252, pp. 69–70
3. K.P. Werrell, The evolution of the cruise missile (Maxwell Air Force Base, Ala, Air University
Press, 1985), p. 8
4. W. Wagner, W.P. Sloan, Fireflies and other UAVs (Arlington, Texas, Aerofax Inc., 1992), IX
5. A.H. Cordesman, A.R. Wagner, The lessons of modern war—vol. IV (The Gulf War, Boulder,
Westview, 1996)
6. C.A. Jones, Unmanned aerial vehicles (UAVs): an assessment of historical operations and
future possibilities, Maxwell AFB, AL, Air Command and Staff College, March 1997, p. 63
7. R. Mkahl, N.S. Moh, M. Wack, Modeling and simulation of standalone photovoltaic charging
stations for electric vehicles, in XIII international conference on computer, communication
8. J.C. Sanna, M. Gawronska, A. Damiano, Energy performance of concentrator and flat plate
photovoltaics in the mediterranean area, in 2015 IEEE international conference on renewable
energy research and applications (ICRERA). (Palermo, Italy, 2015), pp. 760–765, 22–25 Nov
2015
9. M. Steiner, G. Siefer, T. Schmidt, M. Wiesenfarth, F. Dimroth, A.W. Bett, 43 % sunlight to
electricity conversion efficiency using CPV. IEEE J. Photovolt. 6(4), 1020–1024 (2016)
10. S. El Himer, A. Ahaitouf, S. El-yahyaoui, A. Mechaqrane, Performance analysis of S. El Himer,
A. Ahaitouf, Sara El-yahyaoui, Abdellah Mechaqrane, Performance analysis of non-imaging
fresnel lens as a primary stage for CPV units, IOP conference series: earth and environmental
science, p. 012029 (2018). https://doi.org/10.1088/1755-1315/161/1/012029
11. R. Leutz, A. Suzuki, A. Akisawa, T. Kashiwagi, Design of a non-imaging fresnel lens for solar
concentrators. Solar Energy 65, 379–387 (1999). https://doi.org/10.1016/b978-008043895-5/
50228-9
12. M. Akisawa, K. Hiramatsu, Ozaki, design of dome-shaped non-imaging fresnel lenses 3 taking
chromatic aberration into account. Sol. Energy 86, 877–885 (2012). https://doi.org/10.1016/j.
solener.2011.12.017
13. T.T. Pham, N.H. Vu, S. Shin, Design of curved fresnel lens with high performance creating
competitive price concentrator photovoltaic. Energy Procedia 144, 16–32 (2018). https://doi.
org/10.1016/j.egypro.2018.06.004
14. F. Languy, C. Lenaerts, J. Loicq, T. Thibert, S. Habraken, Performance of solar concentrator
made of an achromatic Fresnel doublet measured with a continuous solar simulator and comparison with a singlet. Solar Energy Mater. Solar Cells, 70–76 (2013). https://doi.org/10.1016/j.
solmat.2012.10.008
288
S. E. Himer et al.
15. G. Vallerotto, M. Victoria, S. Askins, I. Antón, G. Sala, Experimental characterization of
achromatic doublet on glass (ADG) fresnel lenses. AIP Conf. Proc. 1881, 030010 (2017).
https://doi.org/10.1063/1.5001421
16. Z. Zhuang, F. Yu, Optimization design of hybrid Fresnel-based concentrator for generating
uniformity irradiance with the broad solar spectrum. Opt. Laser Technol. 60, 27 (2014). https://
doi.org/10.1016/j.optlastec.2013.12.021
17. G. Wang, Z. Chen, P. Hu, X.F. Cheng, Design and optical analysis of the band-focus fresnel
lens solar concentrator. Appl. Therm. Eng. 102, 695–700 (2016). https://doi.org/10.1016/j.app
lthermaleng.2016.04.030
18. K. Ryu, J.G. Rhee, K.M. Parkc, J. Kim, concept and design of modular fresnel lenses for
concentration solar PV system. Sol. Energy 80, 1580–1587 (2006). https://doi.org/10.1016/j.
solener.2005.12.006
19. T.T. Pham, N.H. Vu, S. Shin, Novel Design of primary optical elements based on a linear
fresnel lens for concentrator photovoltaic technology. Energies 12, 1209 (2019). https://doi.
org/10.3390/en12071209
20. K.K. Chong, F.L. Siaw, C.W. Wong, G.S. Wong, Design and construction of non-imaging
planar concentrator for7 concentrator photovoltaic system. Renew. Energy 34, 1364–1370
(2009). https://doi.org/10.1016/j.renene.2008.09.001
21. M. Tian, Y.H. Su, H. FeiZheng, G. Pei, G. Li, S. Riffat, A review on the recent research
progress in the compound parabolic concentrator (CPC) for solar energy applications. Renew.
Sustainable Energy Rev. 82, 1272–1296 (2018). https://doi.org/10.1016/j.rser.2017.09.050
22. Y. Yu, N. Liu, G. Li, R. Tang, Performance comparison of CPCs with and without exit angle
restriction for concentrating radiation on solar cells, angle restriction for concentrating radiation
on solar cells. Appl. Energy 155, 284–293 (2015). https://doi.org/10.1016/j.apenergy.2015.
06.012
23. J.P.F. Rodríguez, H. Baig, E.F. Fernández, F. Almonacid, T. Mallick, P.P. Higueras, Optical
modeling of four fresnel-based high-CPV units. Sol. Energy 155, 805–815 (2017). https://doi.
org/10.1016/j.solener.2017.07.027
24. J.P.F. Rodríguez, E.F. Fernández, H. Baig, F. Almonacid, T. Mallick, P.P. Higueras, Development, indoor characterisation and comparison to optical modelling of four fresnel-based
high-CPV units equipped with refractive secondary optics. Sol. Energy Mater. Sol. Cells 186,
273–283 (2018). https://doi.org/10.1016/j.solmat.2018.06.050
25. S. El Himer, A. Ahaitouf, S. El-yahyaoui, A. Mechaqrane, Parametric optimization and performances analysis of four secondary optical elements for concentrator photovoltaic systems. Int.
J. Renew. Energy Res. IJRER 8, 4 (2018). https://doi.org/10.1063/1.5053502
26. S. Lokeswaran, T.K. Mallick, K.S. Reddy, Design and analysis of dense array CPV receiver
for square parabolic dish system with CPC array as secondary concentrator. Sol. Energy 199,
782–795 (2020). https://doi.org/10.1016/j.solener.2020.02.075
27. F. Trespidi, G. Timò, Optical design of an off-axis mirror based solar concentrator, in AIP
conference proceedings 1881, p. 030008 (2017). https://doi.org/10.1063/1.5001419
28. F. Trespidi, E. Malvisi, R. Parmesani, Shape characterization of the primary mirror of a mirror
based solar concentrator, in AIP conference proceedings 1881, p. 030009 (2017). https://doi.
org/10.1063/1.5001420
29. J.P.F. Rodríguez, E.F. Fernández, Á.F. Solas, F. Almonacid, D.L. Talavera, P.P. Higueras, Optimization of an ultra-high CPV Cassegrain-Koehler unit with 2000_ concentration ratio, in AIP
conference proceedings 2149, p. 70004 (2019). https://doi.org/10.1016/j.enbuild.2017.01.048
30. J.L. Cruz-Campa et al., Microlens rapid prototyping technique with capability for wide variation
in lens diameter and focal length. Microelectron. Eng. 87(11), 2376–2381 (2010)
31. J. Yoon et al., Ultrathin silicon solar microcells for semitransparent, mechanically flexible and
microconcentrator module designs. Nat. Mater. 7(11), 907–915 (2008)
32. C. Domínguez, et al., A review of the promises and challenges of micro-concentrator
photovoltaics. AIP Conf. Proc., vol. 1881 (2017)
33. N. Hayashi et al., Thin concentrator photovoltaic module with micro-solar cells which are
mounted by self-align method using surface tension of melted solder. AIP Conf. Proc., vol.
1881 (2017)
General Parametric of Two Micro-Concentrator …
289
34. A. Ritou, Développement, fabrication et caractérisation de modules photovoltaïques à concentration à ultra haut rendement à base de micro-concentrateurs. PhD thesis Grenoble Alpes
University (2016)
35. B.H. Jared et al, Micro-concentrators for a microsystems-enabled photovoltaic system. Opt.
Soc. Am. 22 (2014)
36. Baranov, V.K. Parabolo-toroidal mirrors as elements of solar energy concentrators. Appl. Sol.
Energy (USSR) (Engl. Transl.) (U. S.), 3, pp. 11–14 (1996)
37. V.M. Andreev, A.B. Kazantsev, V.P. Khvostikov, E.V. Paleeva, V.D. Rumyantsev, M.Z. Shvarts,
High-efficiency (24.6% AM 0) LPE grown AlGaAs/GaAs concentrator solar cells and modules,
in Proceedings of the 1994 IEEE 1st world conference on photovoltaic energy conversionWCPEC (A joint conference of PVSC, PVSEC and PSEC) (Waikoloa, HI, USA), 5–9 Dec
1994
Download