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