Machine learning techniques in internet of UAVs for smart cities applications

Abstract

Recently, there were much interest in technology which has emerged greatly to the development of smart unmanned systems. Internet of UAV (IoUAV) enables an unmanned aerial vehicle (UAV) to connect with public network, and cooperate with the neighboring environment. It also enables UAV to argument information and gather data about others UAV and infrastructures. Applications related to smart UAV and IoUAV systems are facing many impairments issues. The challenges are related to UAV cloud network, big data processing, energy efficiency in IoUAV, and efficient communication between a large amount of different UAV types, in addition to optimum decisions for intelligence. Artificial Intelligence (AI) technologies such as Machine Learning (ML) mechanisms enable to archives intelligent behavior for unmanned systems. Moreover, it provides a smart solution to enhance IoUAV network efficiency. Decisions in data processing are considered one of the most problematic issues related to UAV especially for the operations related to cloud and fog based network levels. ML enables to resolve some of these issues and optimize the Quality of UAV network experience (QoE). The paper provides theoretical fundamentals for ML models and algorithms for IoUAV applications and recently related works, in addition to future trends.

Keywords

IoUAV machine learning deep learning QoE network optimization smart unmanned systems

1 Introduction

Nowadays, environmental sensor networks offer a new model with smart behavior that enables them to adapt to nearby environment and energy accessibility, and usages non-standard radio networks for connections that allow wireless trunks to send sensor data to the Internet. In an IoT environment, smart unmanned aerial vehicle (UAV) can have an internet connection permitting them to transmit information directly to ground stations and users to cooperate with them simply and consistently [1]. IoT applications within the UAVs environment can be separated into two fundamental categories of administrations, they are natural asset administration and natural security. The concept of asset administration is connected to characteristic assets and categories as natural, non-biological and renewable assets, incorporates creatures, people, woodlands, feathered creatures, water, metal and etc. [2]. All of these assets are likely to diminish essentially or be influenced by a few components, counting contamination, squander and misuse. For these factors, unmanned aerial vehicle (UAV) provides an important application for monitoring and surveying. UAVs need fast and effective management techniques, in addition to IoT technology which can provide an effective way to communicate UAV related to the environmental resources with monitoring and observation centers and research and to take an appropriate decision in utilization and consumption of these resources [3].

AI based UAV-to-everything (V2X) network acquires information from various unmanned nodes i.e., cars, trains, buses, and etc., allows to growth realization of driver and forecasts to achieve intelligent behaviors. This evolution has focused to chance of understanding smart driving which constructed on the idea of repetition actual driving comportment. The real driving behavior artificial learning helps to avoid human faults and bring safety, contented to driver. Internet of UAVs is discussed to modify the collaboration between the UAV, ground base-stations, on flying station and environment [4]. IoT based AI technologies enables UAV to send information and real application data between different UAVs networks and provides several applications related to products delivery, environment monitoring, military and for search and rescue as presented in Fig. 1.

Fig. 1

UAV Applications.

Machine learnings (MLs) algorithms are accountable for widespread range of artificial intelligent (AI) applications [5]. There are many ML techniques used as methods for different AI solutions, which provide an effective representation of unlabeled data. ML also provides different solutions such as collaborative optimization of energy consumption for UAV, optimum route prediction of electric UAV and minimize in traffic congestions. Moreover, the use of ML in IoUAV promises to solve different related issues such as, traffic prediction based path, data routing, vehicular block chain, congestion, load balancing, cyber-physical attack mitigation, resource management based energy efficiency [6]. Other considerations of ML-based IoUAV applications are related to the serious problematic issues for IoUAV through real-time applications. They able to improve the quality-of-experiences (QoEs) in critical application. ML-based methods have completely altered the vision of IoUAVs [7]. It allows the portable nodes for sending real-time applications and contents in IoUAV system, to the end-users in their corresponding field. Due to the importance of ML mechanisms in the IoUAV applications, and their achievements to enhance different IoUAV aspects. This paper discussed an extensive detail about ML and deep learning methods in IoUAV, in addition to the possibility of its use in several IoUAV applications.

2 Related works and motivation

Recently, many scientific researches were presented in the field of in wireless communication technology that are related to the smart transportation and vehicle communication management. Also, the emergence of a new generation of unmanned aerial vehicle (UAV) contributed to the tremendous development of what is known as the Internet of unmanned aerial vehicle systems [8]. The increasing interest in the use of internet in UAV has led to the emergence of a new type of vehicle networks which facing a various problematic issue related to the management of resources and quality of service and experience, in addition to interaction with the surrounding environment of vehicles. In ground vehicular ad hoc network (VANET), the communication take place between vehicle-to-vehicle (V2V), UAV-to-U2V. In UAVs networks, communications occurred between the aerial base station (ABS) and ground base stations (GBSs) [9]. The GBSs enables to transfer the information between the UAVs. Hybrid UAV network architecture enables to support communications and interactions between cellular communication and UAVs to provide voice services.

IoUAV provides an internet access to the UAVs in flying environment. Its enables to interconnect the UAVs ground base stations networks to various wireless networks technologies i.e., MOST/WiFi for V2S, LTE/V2I/V2X, IEEE/WAVE for V2V/V2R. IoUAV provides different applications related to the intelligent flying transportation system, for traffic monitor in UAV environment. The enhancement of UAV transportation system are related to the ability of accident reduction and traffic roads improvement. Artificial intelligence techniques have tangible impacts on many issues related to the IoUAV networks [10]. Such issues are the ability of special decisions and forecast for different aspects related to traffic management and monitor, bigdata processing, resources management and energy, intellectual collaboration with nodes to afford high-quality service. There are many recent studies on how to use artificial intelligence techniques such as machine learning in smart UAV applications. Machine learning is used to invent new methods for of the challenge related for improving smart decisions in IoUAV applications for various issues and related aspects [11]. It is helpful to give a comprehensive presentation to study some concepts about ML use in IoUAV, as well as to explain research areas that could enhance the progression of these types of network. This is one of the main motivations for which this paper is written. Table 1 summaries the recently studies related to the AI and ML algorithms used in the UAVs applications.

Table 1
Summarization of AI and ML Approaches in IoUAVs Applications

Approach Feature Advantage Citation

ML and Faster region-based convolutional neural network (Faster R-CNN) for Navigation Autonomous navigation of a small UAV between trees Accurate and robust for autonomous navigation Lee et al., 2021 [102]

Machine learning for high data rates of UAV in 5 G Real-time object detection and activity tracking based on 5 G cloud Remotely control the UAVs A. Ahlawat et al., 2020 [106]

Deep learning to medical field application Optimize the UAV biomedical engineering technology systems Optimal performance for realistic medical situations Hao Meng et al., 2020 [105]

Multi-agent deep Q learning (DQL) for UAV trajectory planning Optimize throughput for 3D path design with constraint of flight speed of UAVs Improves the performance gain in terms of minimum throughput maximization JIE TANG et al., 2020 [103]

Machine learning to enhance the capabilities of UAVs Reduce UAVs challenges and enhance the real time monitoring Enhance data collection and processing Asharul et al., 2019 [98]

Reinforcement learning for distributed sense protocol UAVs trajectory control and resource management Faster sensory data received by ground base station Jingzhi et al., 2019 [99]

Game theory and reinforcement learning for UAVs data offloading process Test offloaded data in UAVs and selected MEC servers according Best response UAVs dynamics framework Nicholas, 2019 [101]

SPF learning algorithm to speed up UAVs the data processing scalable the computational power with web data analytics algorithm Increase UAV utilization E. Bijolin et al., 2019 [104]

ML and ANN technology for security challenges in UAVs adaptive UAVs system resources for secure operations Secure real-time multimedia streaming Ursula et al., 2018 [100]

Approach	Feature	Advantage	Citation
ML and Faster region-based convolutional neural network (Faster R-CNN) for Navigation	Autonomous navigation of a small UAV between trees	Accurate and robust for autonomous navigation	Lee et al., 2021 [102]
Machine learning for high data rates of UAV in 5 G	Real-time object detection and activity tracking based on 5 G cloud	Remotely control the UAVs	A. Ahlawat et al., 2020 [106]
Deep learning to medical field application	Optimize the UAV biomedical engineering technology systems	Optimal performance for realistic medical situations	Hao Meng et al., 2020 [105]
Multi-agent deep Q learning (DQL) for UAV trajectory planning	Optimize throughput for 3D path design with constraint of flight speed of UAVs	Improves the performance gain in terms of minimum throughput maximization	JIE TANG et al., 2020 [103]
Machine learning to enhance the capabilities of UAVs	Reduce UAVs challenges and enhance the real time monitoring	Enhance data collection and processing	Asharul et al., 2019 [98]
Reinforcement learning for distributed sense protocol	UAVs trajectory control and resource management	Faster sensory data received by ground base station	Jingzhi et al., 2019 [99]
Game theory and reinforcement learning for UAVs data offloading process	Test offloaded data in UAVs and selected MEC servers according	Best response UAVs dynamics framework	Nicholas, 2019 [101]
SPF learning algorithm to speed up UAVs the data processing	scalable the computational power with web data analytics algorithm	Increase UAV utilization	E. Bijolin et al., 2019 [104]
ML and ANN technology for security challenges in UAVs	adaptive UAVs system resources for secure operations	Secure real-time multimedia streaming	Ursula et al., 2018 [100]

For security purposes in cellular-connected unmanned aerial vehicles (UAVs), [100] reviews the goal of ML technology for security challenges related to UAV-based delivery systems. They present the artificial neural network (ANN) for UAV based real time applications, and UAV empowered smart transportation system. The use of ANN and ML enable to adaptive the UAVs system resources and guarantee secure operation. The proposed approach shows the benefits of ML and ANN for of the aforementioned cellular-connected UAV application use case. Study presented by [98] touched on the importance of machine learning models in enhancing the capabilities of unmanned aerial vehicles. They present that ML will enhance the UAVs capabilities. The study shows the ML ability in reducing many challenges related to UAVs and the contribution to enhance the real time monitoring. The authors also present the ability of ML in enchanting data collection and processing, in addition to the prediction in different UAVs applications. In [99], study proposed a new distributed sense and transmission protocol to coordinate the UAVs. Authors show that, the use of reinforcement learning in cellular Internet of UAVs will helps to solve problems related to trajectory control and resource management. Authors evaluate the proposed protocol as a multi-UAV Q-learning algorithm, and the results show faster and to a higher reward for valid sensory data received by ground base station.

For intelligent UAV transportation system, traffic monitoring and balancing are two important for improves the follow of traffic. In [101] author presents an AI technology for UAVs data offloading process in a multi-server mobile edge computing (MEC) environment. The proposed study uses the game theory and reinforcement learning to determine the offloaded data in UAVs and selected Mobile edge computing (MEC) servers according. The overall framework performance evaluation is achieved via modeling and simulation, in terms of its efficiency and effectiveness, under different operation approaches and scenarios which gives best response dynamics framework. Authors in [104] presents UAVs data analytics methods with high efficiency to analyses the data for making key decisions are being identified. They review the use of 3D Path planning algorithm using visibility graph proposed to know the shortest path and the amount of data collected under visibility graph method. The study uses the SPF learning algorithm to speed up the data processing. In addition to expanding data into the cloud with bandwidth management. The study also focusses on scalable the computational power with web data analytics algorithm for end process automation [104]. The performance of the model is increased the utilization. The study in [105] reviews a detailed description of the UAVs network applicability to achieve close optimal performance and efficiency of medical engineered devices. The study also provides the use of deep learning to medical field application. They present a details about the multi armed bandit (MAB) approach in solving the UAV biomedical engineering technology device, in addition to optimized utilization of biomedical engineering technology systems. The conclusion shows deep learning achieves optimal performance for realistic medical situations.

Authors in [106], proposed a real-time applications of machine learning for high data transfer rates of UAV in 5 G. The authors proposed the ML for real-time object detection and activity tracking using a UAV mounted system. The study depends on data transmission between the UAVs and cloud by ML to execute on the Cloud in real-time through the cellular network. The proposed method enables the UAVs to be controlled remotely from the virtually anywhere, without physical presence of a pilot. An investigation on joint unmanned aerial vehicle (UAV) trajectory planning and time resource allocation in a multiple UAV has been reviewed in [103]. Authors proposed a model for throughput optimization involves joint optimization of 3D path design and channel resource assignment with the constraint of flight speed of UAVs and uplink transmit power of IoT devices. The study uses the multi-agent deep Q learning (DQL) strategy for optimization. The simulated algorithm improves the performance gain in terms of minimum throughput maximization. Researchers in [102] provide a novel approach to the autonomous navigation of a small UAV between trees using a single camera. The study uses the ML and faster region-based convolutional neural network (Faster R-CNN), as a control strategy to avoid collision with tree. Detection are based on the captured tree images and analyses the heights, distances from the UAV in addition to find the widest obstacle-free space. The control strategy allows the UAV to navigate until any approaching obstacle is detected and to turn to the safest area before continuing its flight. The results of proposed approach show an accurate and robust for autonomous navigation between trees.

3 Machine learning approach

Machine learning (ML) is widely used in most applications related to decision making and resource management due to its great ability in analyzing large data sets and building decision systems. ML offers a number of unique solutions that fall within the context of the IoUAV like autonomous driving, speech recognition, and consumer behavior prediction [12]. However, ML requires a long training period to achieve proven and reliable outputs, so the use of distributed systems has been concerned to achieves parallelization and total amount to I/O bandwidth enhancements. In addition, the use of distributed machine learning systems reduces the pressure on storage in the case of dealing with big data processing. According to the aforementioned, the use of developed algorithms that allow parallel computations, data distribution, and flexibility to failure helps a lot in dealing with large data-volume training. In general, ML method are divided into two phases, training phase and prediction phase (see Fig. 2). The training phase includes feeding a huge sum of training information to allow ML model to train on using ML algorithms. according to this, to selects the better ML algorithm for specific application, the considerations should be focused on the purpose of the system application and relationship with the set of algorithm hyper parameters. The second phase is prediction which is used to position the trained model in practice [13]. The trained model receives new data as input and gives an output is a prediction.

Fig. 2

General Machine Learning Concept.

Two approaches are considered in applying the concept of distributed machine learning parallelism, data parallel and model parallel (see Fig. 3). The data parallel approach partitions the same data amount as of working nodes system applying the same algorithm to different datasets [13]. For the parallel model approach, an exact copy of the entire data sets is processed by the nodes.

Fig. 3

Parallelism in Distributed Machine Learning.

In UAVs applications, ML enables data to be classified and evaluated the related values to learn how to make a decision. Machine learning techniques can find the type of UAVs from their characteristics, where the algorithm can be provided with a set of training data that contains all the characteristics of the UAVs, and then obtain a comprehensive description of the type of UAVs. This method can be used for this application in UAVs detection systems [14, 15]. ML enables to reduce power consumption when moving and changing positions by predicts wireless data traffic. This mechanism helps to determine the mobility to reduce power for UAVS. in addition, ML is also used in predicting the channel in order to reconstruct the radio map. In channel modeling for complex UAV-to-UAV communications and ground-to-UAV channels, ML enable to predicts UAV-to-UAV path loss [37]. It also helps to evaluates predictions about path loss from several parameters such as propagation distance, and UAVs elevation.

4 Intelligence IoUAV network

In IoUAV network architecture, the problematic issues associated with the layers that accountable for presentations and functionals process in IoUAV can be resolve by the AI technologies. The virtual cloud infrastructure enables to store IoUAV network information, and process it, in addition to make decisions according to analyzed information [14]. The computations and analysis in IoUAV are depends on the bigdata analysis (BDAs) aerial vehicular-cloud-computing (VCCs) systems. IoUAV cloud based computing requires an intelligent service management to provides different applications and services related to safety traffic management. AI procedures with data traffic enabled by the IoUAV network cloud servers, to perform an intelligent decision making for customer services. Since UAVs are able to communicate with the Internet, it is possible that the Io UAV network may contain cybersecurity vulnerabilities [15]. The vehicles cyber-physical systems (VCPSs) are used as a IoUAV vehicles networks design to distribute traffics via Internet access. AI based VCPS able to processing the big data and secure the traffic and computations in fogs and clouds.

Caching operations and edge computing are mostly concerned in IoUAV networks, which are related to various factors. AI provides an optimization approach for caching and edge computing to enhance these parameters, i.e., adaptive topologies, and resources assignments and allocations management (RAAM), in addition to channel conditions [16]. Machine learning algorithms able to provide and intelligent interactions between the IoUAV and its environment, which will enhance the overall IoUAV network utilization based on the collected environmental data. One of the mostly used ML algorithms are deep neural network(DNN) and Q-learning, which they able to provide an optimized decision making in some IoUAV applications according to the learned resources. In the IoUAV network architecture, Q-learning and DNN are engaged with the detached AI layer which is responsible for bigdata analysis and cloud computing [17]. The separated AI layer operates different tasks such as storing, data processing and analysis, coordinate with other IoUAV architecture layers, in addition to make decisions with respect to the status of IoUAV network.

4.1 Intelligent IoUAV communication

Multimedia communications in IoUAV enables to exchanges data between onboard UAVs devices. IoUAV based multimedia commination network integrates different parties such as IoT modules, personal networks, sensors networks and other IoUAV drivers. Data transmission scalability and flexibility are important paradigms when considering the integrations between humans, machines and UAV system [18]. The integration enables to improve safety traffic for UAV, enhance the IoUAV data traffic and to overall vehicular system efficiency.

In the context of quality, it found that heavy consumption of power and storage capacity, as well as heavy data exchange, IoUAV is likely degrades the IoUAV network quality performance. AI technologies improve the procedures of self-driving in UAVs and reduce the low quality risks. ML algorithms provides solution for quality improvement by developing smart utility to ensure high information traffic in IoUAV, and enable efficient resource monitoring and management as well [19]. In IoUAV, the multimedia services are depending on ability of It is based on the ability to analyze transport information through IoT-based platforms. In Fig. 4, the IoUAV multimedia communications are based on transferring urgent data via adaptive and smart wireless communication. Quality of services (QoS) monitoring is enabled by the UAVs client, which is evaluate the IoUAV traffic and categorize the follow of data according to the conceptual of data sensitivity, pre-store in real-time, and definition resolution.

Fig. 4

IoUAV Multimedia Communication Models.

4.2 Intelligent IoUAV edge computation

Flying Edge computing is a technology provides communication to deliver cloud services from the network edge. In IoUAV, flying mobile edge computing (FMEC) enables to provides communications for critical UAVs applications. AI in FEMC helps to develop an edge information system (EIS) based on machine learning model for IoUAV to achieves intelligent data acquisition and processing [20]. For data processing, ML based edge computing enables to process data at the network edge with low latency requirements for mission-critical and low bandwidth connections. In addition, ML edge based IoUV provides spatial locality for flying route conditions, and map information for traffic conditions. Moreover, ML enables to optimize load balancing and perform offloading computation in order to utilize the IoUAV system [21]. Performance utilization can be ensured through the cooperation between the edge-caching and computing which are depend on the policies limitation in dynamic IoUAV network applications.

AI achieves cognitive edge computing capabilities to the IoUAV network to provide dynamic computing services, and overcome the limitations of caching and computing policies dynamic environment [22]. Moreover, it enables to enhance the efficiency of energy and quality of experience (QoE), in addition to improve IoUAV resource management. IoUAV edge-based AI architecture as shown in Fig. 5 enables to predict the actual behavior of vehicle environment information by cooperating with the surrounding environment conferring to the information on the present status for offload to the edge computing.

Fig. 5

IoUAV Edge Based AI Architecture.

Deep learning (DL) in the IoUAV edge Network enables to intelligently manage the computing resources by relying on collaborative buffering to raise the efficiency of edge computing management [23]. Moreover, DL ensures to enhance the operation of handovers between the IoUAV and RSUs in dynamic network. DL provides an important role in the trade-off between prediction accuracy, temporary buffering, and handover execution, in addition to and intelligently share assigned edge-caching and bandwidths (BW).

4.3 Intelligent IoUAV for everything

IoUAV based on Vehicle-to-everything (V2X) paradigm is a IoUAV network architecture enables communication between different vehicular network types such as vehicle-to-personal devices vehicle-to-vehicle, and vehicle-to-ground base station. Traffic and energy efficiency, in addition to road safety are essentially aspects related to the IoUAV for everything applications. AI provides intelligent solutions to perform tasks related to the IoUAV based V2X, such as traffic congestion, UAV electric charging, in addition to improve the services based on vehicle locations [24, 25]. AI based IoUAV together with V2X can enable intelligent applications for traffic flow prediction and management, autonomous transport facilities, and congestion control. Referring to previous studies in the field of internet of things (IoT), there are different AI methods are widely for optimization and prediction in the vehicle network applications [26]. Techniques such as game learning, swarm intelligence, fuzzy logic, and machine learning are all provide solutions for network planning, scheduling, optimizing and data processing.

In IoUAVs based V2X application, ML provides an intelligent method to enhance the UAV network efficiency and safety for autonomous driving application, as well as for human remote control interface [27]. The use of V2X system enables the UAVs to share and exchange data with road vehicles or walker infrastructure. Another aspect of machine learning capabilities is its ability to predict the exact exchange and flow of traffic in relation to IoUAV. To develop data exchanges between the UAVs and other V2X networks, ML helps to enhance network planning for different UAVs and V2X cluster groups based on location, and data modeling [28]. Moreover, ML enable to balance the energy consumption, in addition to trade-off between mobility and cluster radio coverage, as well as to enhance the QoE.

5 Artificial intelligence for IoUAV QoE optimization

Quality-of-experience (QoE) is a term describe measurements of network knowledge performance and network perception by the user. In IoUAVs, QoE provides an application experience benchmark, which is ensures high quality of data exchanged in network. In IoUAV, low quality of user experience is a critical issue in respect to UAVs applications [29]. QoE evaluates the degree of integration coherent and flexibility between the IoUAV parties, i.e. sensors, humans, and machines. It also provides a measures helps to evaluate the user perception enhancement, and power consumption as well. Power and buffer consumptions based QoE is in IoUAV enables to improve the UAVs transportation and data traffic. QoE in IoUAV provides an optimal solution to balance the quality risk in sensitive applications. In IoUAV, the QoE can be affected by many factors associated with network structure. IoUAVs network routing, vehicles positions, and topology are main factors may impact the network energy efficiency [30].

AI technologies helps to optimize the IoUAVs energy efficiency by providing decisions to operate the network functions. Moreover, AI provides an intelligent solution to enhance the network routing selection based on power and traffic quality, in addition to optimize the energy and buffer consumption during multimedia communication [30, 31]. Quality measurement framework-based machine learning enables to analysis the QoE and evaluates the quality degrading factors according to the collected information from IoUAV network, considering different impact factors related to the IoUAV communications, vehicles energy, and resource resources allocation [32].

One of the most important applications related to IoUAVs is the search and rescue services, which relies on capturing information via video data. Video data traffic in such application requires an intelligent solution to fulfill the requirement of high multimedia communication, ensured by maximum QoE. Buffer allocation-based ML provides an optimized mechanism for QoE optimization. ML algorithms enable to predicts the peak variable rate of multimedia traffic and according, allocates the proper buffer size [33]. In addition, the mechanism should consider to optimize the energy and video rate adaptation as well. From other side, the video contents may affect the video transmission in IoUAV applications, and degrades the QoE. Dynamic code rate adapting helps to enhance the QoE. ML enables to optimize the QoE by evaluate the video processing in pixel level and data dimensions automatically [32, 33]. The processing evaluation is depending on the ability of ML to adapt the buffer allocation and video data rate.

Other QoE consideration is related to energy management in IoUAV. Efficient power resource management during IoUAV communication is depends on the vehicles discharging time [34]. Frequent charging and battery replacements in IoUAV limit appropriate use of aerial vehicles in most applications. ML enables to optimize the peer-to-peer distributed network of vehicles charging stations by energy scheduling and trading between the vehicles and charging station [55].

The alleging schedule scheme should deliberate the optimized QoE in a situation of higher UAV’s movement and restricted area that cover the ground base station [36].

6 Machine learning based IoUAV network

ML approaches have distinctive designs, classification and preparing which have been broadly utilized for forecast issues and brilliantly overseeing. In IoUAV services, machine and reinforcement learnings would give a direct performance to advance versatility and versatility. they could also give way determination or course for optimized IoUAV systems. The employments of machine learning with software-defined networks (SDNs) in IoUAV guarantee minimum delay and maximum throughput for operation and support techniques. ML-based SDNs together would move forward the IoUAV organize execution with steady and prevalent steering administrations [37]. They guarantee ideal directing arrangement adjustment agreeing to the IoUAV environment detecting and learning to attain superior utilization. Figure 6 appears the ML functions can be deployed within the IoUAV systems. Within the space of IoUAV organize security, ML with SDN brings a few one-of-a-kind preferences to the arrangements of security arrangements. For cybersecurity issues, the central management on the programming layers associated with API access would be helpful to create ML applications associated with SDN information planes to send control messages to the application layers upon UAVs services demands [38].

Fig. 6

ML Functions in IoUAV network.

In cognitive Web of UAVs (CIoUAV) applications such as programmed driving, the robotization and network are exceptionally vital in self-driving angles which they ought to be adequate of insights to decrease the street mischances. ML could handle UAVs system control to empower mistake of auto-driver. CIoUAV empowers to send ML based cloud into transportation framework for cybersecurity risks and problems [39]. In CIoUAV control and cognition layers, Machine learning approaches give vital administrations for distinctive work layers i.e., behavior of driving and wellbeing observing, design and feeling examination, in expansion to arrange asset assignment and optimization. To progress driving security and productivity within the IoUAV transportation framework, DL plans afford a smart decision making to assess the vital persuasive collisions likelihood components and chance of conceivable mischances within the IoUAV [40]. There are diverse DL procedures could be utilized for collisions forecast and mishap determining i.e., fuzzy logics and support vector machines (SVMs), neural-networks (NN), generic algorithms (GA).

A. ML for IoUAV Edge-caching

The operational fabulousness and taken a toll productivity in IoUAV are depending on caching and computing plan. To effectively move forward the QoS in IoUAV application, edge-caching situations and computing offload at UAVs and the ground base station (GBS) can guarantee to ensure productive QoS. Machine learning gives plans to handle issues experienced in computing, communication and caching for IoUAV. Numerous MLs plan could be utilized for edge-caching in IoUAV [41]. Directed learning gives generally great caching choices, in expansion to IoUAV activity levels classification, expectation and substance request. Non-supervised learning could be connected to edge-caching plan by cluster number of UAVs in to distinctive bunches agreeing to the behavior and information data history [42]. The ML-based clustering conspire can anticipate the information request depends on the interface or social relations of the whole UAVs bunch.

The support learning conspire like Q-learning method, will empower to disperse cache substitution methodology concurring to the substance notoriety. Additionally, it can gauge the obscure notoriety of caching substance. Coordinates mobile edge computing (MEC) servers in IoUAV arrange will offer assistance to diminish the workload at ground stations. It too empowers to create UAV system to perform offloading computation amid to develop an edge-based intelligence for edge-caching (see Fig. 7). The utilize of significant Q learning would optimize the parameters of computing and caching for asset assignment. Q-learning would decide the activities from the gathered information and status of MEC and GBS servers, in expansion to each UAV’s versatility, channel data, caching substance and computing [42, 43]. These activities are sent to UAVs. Deep Q learning will select the leading set of caching activity for RSU, MEC and UAVs to assist the inquires and to calculate the offload assignments.

Fig. 7

IoUAV Edge Based Caching Computational Scenario.

ML integration with edge-caching have problematic issues associated to information preparing and examination. The dissemination and tall thickness of information are vital issues for training and learning preparations. In expansion, inadequately of computing assets in to control the tall dimensional data that cannot give exact buffering choices [44]. To unequivocally agreeable ML at the IoUAV organize edge to upgrade the keen obligations of the edge, it requires a compelling learning approaches for enormous tall dimensional data that be built up in arrange to offer exact estimate of the saved data at the IoUAV organize edge. Besides, ML plans arrangement in IoUAV services will extricate considerable delicate and basic data, and in case any spillage of data, it could cause genuine secrecy, protection, and security issues [45]. For these concerns, edge-caching framework must be secured by security and security protecting plans and ought to be created totally unlike framework layers, i.e., data processing, transmission, capacity levels for both edge organize and UAVs, and data access.

B. Deep Learning for IoUAV Edge Computing

The implementation of computing intensive applications on resource constrained UAVs still encounters challenges associated to offload IoUAV framework. RL will give a shrewdly offload framework for vehicular edge-computing [46]. The combination of RL with vehicular edge-computing offer assistance to plan offload demands, in expansion to apportion IoUAV arrange assets. RL optimize the planning and asset assignment in IoUAV to maximize the QoE. In IoUAV, an unmanned aerial vehicle (UAV) calculates utility values, and passed the offloading demands to the ground stations [47]. The stations perform assignment planning and asset assignment. MEC servers empower to get all UAVs offloading errands to execute offload computation. The RL calculations offer assistance to enhance the offload choice by cleverly assignment planning. Figure 8 illustrates architecture of agent-enabled task offloading for UAVs systems. Offloading demands planned within the assignment agreeing to action-value work Q. MEC servers is chosen by UAVs from the accessible getting to list with likelihood and biggest Q value of the present action value work [48]. The utilize IoUAV RL offloading computation would be ensured by guaranteeing agreeable offloading in IoUAV arrange, which can enhance the QoE of UAVs.

Fig. 8

Architecture of Agent-Enabled Task Offloading for UAVs Systems.

In IoUAV offload computing parameters are associated to the offload proportion for each assignment. The UAVs configuration values imperatives are associated to the confinements of UAVs processors and memories asset. Offload optimization is depending on how to play down the work errand idleness and vitality costs. Taken a toll work can be expressed as takes after, $F = \sum_{i = 1}^{N} (L_{i} + w_{i} E_{i})$ (1)

Where, F is a function, L_i denoted for latency cost, N is the tasks number, and E_i forenergy costs is w_i signified for weight proportion between inactivity and vitality fetched. Each offload choice depends on asset time unit space. This implies, the stream of planning errands could be an arrangement in time. For grouping of the N assignments arrives amid a restricted watching time L_obs, c fetched work F can be expressed as,

$F = \sum_{t = 1}^{L_{obs}} (γ^{t - 1} R (t))$ (2)

R(t) denotes as remunerating of t time unit space and γ indicated for compensate markdown proportion to portray warmth of rewards of long run time space on by and large taken a toll work. This add up to fetched work can be utilized to create arrangement organize preparing. Neural networks will give an approach for map the IoUAV surrounded states to the possibilities of activities to be involved in the ML process [49]. The arrangement organizes based RL preparing will accomplish an optimize calculating by means of the compensate esteem of each time opening [50]. This would reduce the weighted entirety of offload inactivity and control utilization fetched and guarantees offloading choice optimization.

In IoUAV based edge-computing, UAVs would perform as a client interface over the edge-computing in MEC server without getting to farther cloud. In this situation, offloading choice for heterogeneous resources is classified as multifaceted process [51]. This since, the management of vehicular edge-computing is changing each time and requires that offload decision ought to be re-calculated, which is leads to noticeable delay in administrations procedure. In expansion, for vehicular benefit, the errand execution advance cannot ensure reasonableness offload lining. Deep RL gives a special choice calculation to realize smart vehicular controlled administrations based on edge-computing framework [52]. It makes a difference to memorize the benefit offloading information, and the perception capacities related to environment information of vehicular versatility and the edge-computing hubs. The offload decision scenario is prepared at the capable edge computing hubs and conveys the choice data to the UAVs for administrations offload. Amid choice show preparing, UAVs sends the IoUAV coefficients to the ground station edge-computing center for overhauling the essential offload decision procedure [53].

C. ML for Dynamic IoUAV Mobility

IoUAV systems may have flow highlights in numerous perspectives i.e., activity, structure, and remote proliferation channel due to its instability. An effective learn and energetic forecast are required to supply an optimization degree in leading, activity stack, and for help the channel estimate module to track channel varieties [54]. ML strategies lead to way better comes about for designing the energetic changes of vehicular channels in expansion to optimize UAV directing and activity stream. ML framework coordinates into RSUs offer assistance to assess activity designs by collecting data approximately UAVs. ML can give cleverly IoUAV steering convention with basic data for profoundly energetic environment [55]. The expectation is dependent on the data sent from the UAV when it roams from GBS to another where is able offer assistance GBS to empower appraise the activity streams.

D. ML for IoUAV Decision Making

In later a long time, independent vehicle (AV) development creates a novel propensity to execute a few cleverly approaches and strategies to upgrade the productive and quality of versatile decision-making. The combination of AI, ML, RI and IoUAV offers tall productive control frameworks that can abused in different applications to suit more versatile, programmed and strong inserted frameworks [56]. Choice making in IoUAV systems requires shrewdly calculations to handle the forms related to driving environment recognition, way arranging, procedure arrange control, and asset administration. For brilliantly driving UAVs framework, a module coordinating the way, behavior, and movement arranging, is required to function in profoundly optimized decision-making calculation. In expansion, decision-making procedures should be taken under consideration the operation of UAVs network. It should be capable to anticipate and study the data associated to UAV stage deficiencies, direction and vitality [57]. These contemplations are bargain with UAVs stage as appeared in Fig. 9. For cognitive-driving decision-making, positioning, semantic understandings, and sensors combination are more contributed within the decision making procedure.

Fig. 9

Intelligent driving an unmanned aerial vehicles (UAV) decision making framework.

Moreover, the shrewdly UAVs and IoUAV frameworks applications confront the decision-making challenges related with collecting and conveying IoUAV huge information to UAVs and interested clients with the point of upgrading street insights encounter. In expansion, making choices related to activity overseeing, street clog and security [58]. Tremendous volumes of enormous information require a more effective and shrewdly component in decision-making methods to decrease street clog, and make strides activity operations. Additionally, the shrewdly decision-making empowers to urge drive the challenges related to viable communication interface between diverse sorts of UAVs and shrewd gadgets, security and protection issues [59]. Numerous machine learning strategies can utilize to contribute in fathoming the over issues. These strategies empower to demonstrate channel totally various IoUAV arrange scenarios. In expansion, it gives perceptively arrangements to maintain a strategic distance from street mishaps by perceptive learning and examination of the drive perceptive utilizing the information gathered from the other UAVs. Subsequently IoUAV systems are inquisitive about exchange packets all over and exchange states among UAVs [60]. ML-based shrewd asset administration for IoUAV systems has ended up greatly critical to create choices on the arrangement of association strategy of control, determination, and asset assignment and task.

E. ML for IoUAV Network Control

Higher IoUAV arrange execution requests efficient arrangements for arrange optimization, adaptation and operation. ML in arrange space would use the capable of ML capacities for modern arrange administration in IoUAV services. The abilities of ML would provide a productive method for interruption location and execution forecast. In expansion, ML empowers the IoUAV arrange to create cleverly choices for organize planning and adjustment depends on organize characteristic and environment [61]. ML calculations will encourage IoUAV organize to classify and foresee of activity designs and organize state. In common, the utilize of ML in optimization systems guarantee numerous arrangements for diverse organizing viewpoints, i.e., information gathering and investigation, cluster decision-making and forecast, demonstrate development an approval, in expansion for organize sending and impedances as appeared in Fig. 10. Since of the IoUAV characteristics depends on the web, information and activity forecast, examination and classification are most vital viewpoints related to IoUAV organize control [62].

Fig. 10

ML for IoUAV Network Control Cycle.

Information gathering and investigation are associated to the steps of gathering a huge sum of agent arrange information, and capacity of characterize the arrange variables. Based on the IoUAV services, information gathering can be assembled from distinctive arrange layers [63]. Concurring to the IoUAV organize states, the offline information collection with tall quality is required for information examination, whereas online information gathering will empower learn the organize execution and adjustment [64]. For IoUAV basic applications, information investigation must discover an appropriate networks behaviour i.e., to anticipate the finest arrange activity execution by analyzing the verifiable information. For information gathering and investigation, it’s imperative to prepare network information by normalization, discretization, and lost esteem completion. ML may be a great choice to assist extricate the organize highlight. For IoUAV systems, ML plays an imperative part in activity forecast and arrange administration [65]. Precision in activity volume estimation in IoUAV systems is considered one of the most variables affect the execution examination of arrange operations, i.e., asset allotment, organize directing, clog and information gushing control. Numerous thinks about attempt to decrease the taken a toll of activity estimation by utilizing ML calculations [64 65]. Activity classification speaks to the require of IoUAV organize applications to be coordinated with web activity stream.

In IoUAV, the web activity classification is critical angles to supply productive organize quality of benefit and quality of involvement. In addition, within the arranged edges, precise internet-of-UAV data flow categorization on could be vital issue and a critical problem of the arrange network environment. In this instance, the imperative of arrange activity classification is to distinguish the UAV arrange applications and control the activity stream as required with in adjusting esteem or in need over each other. In security issues, activity categorization gives a implies of interruptions and pernicious assault’s location [66]. The utilize of ML based on factual highlights will gives a classification situation to more reasonable circumstance for IoUAV organize activity for arrange control and security. In addition, it accomplishes proficiency, versatility, and execution upgrade.

Other contemplations rerated to organize control are arranging activity observing and administration. In IoUAV organize, to guarantee effective arrange optimization, ML empower to adjust the energetic web activity in IoUAV and exploit the QoE/QoS without compromising end-users encounters [67]. ML gives an adjustment of genuine time arrange situations and exploits the client encounter. DL can offer assistance to avoid the inadequacy of conventional TCP protocol procedures by categorizing a parcel misfortune due to patten or connection errors. By ML approaches, it’ll be simple to customized best suited blockage control plot that able to adjust to organize special requirements. ML can methodically prospect critical data from information held by an unmanned aerial vehicle (UAV) and naturally distinguish exceptionally complex joins, permitting UAVs to intellectuals screen their environment, and utilize information for preparing purposes [68]. ML empower to foresee and adjust to the advancement of natural highlights, counting remote channel flow and activity and portability designs, in expansion to design the organize, which gives incredible plausibility to control and oversee the organize data flow. Other DL arrangements associated to creating exact channels model in numerous situations and lessening way misfortune.

These arrangements lie in foreseeing IoUAV topology, and treating serious impedances from other IoUAV utilizing route information and UAVs network. In IoUAV applications, Web activity may be affected by the shortcoming of remote communications [69]. ML advances are able to survey remote conditions without the required for an expansive sum of information sets and by utilizing ANNs strategies, an receive signal strength (RSS) expectation can be achieved in an IoUAV environment.

7 Machine learning based IoUAV applications

ML contributes in numerous IoUAV applications related to rising message transmission for road security and unsafe exercises. In expansion, ML gives modern savvy arrangements for IoUAV administrations and amusement. In arrange to play down the generally vitality utilization of the computational offices and UAVs whereas fulfilling the delay imperative for activity offloading [70]. The utilize of ML innovation in information mining, design acknowledgment, preparing, cognitive computing, is an alternative for choice making, which is able open unused openings for brilliantly IoUAV systems i.e., in driver security, savvy transportation and independent driving applications.

Machine learning plays a crucial part in an unmanned aerial vehicle (UAV) cleverly driving application, which is make UAVs able to discernment and estimation to proficiently oversee the UAV driving framework [71]. ML makes the UAVs self-automated which can move forward society by diminishing street mishaps. In common, the self-driving UAVs are exceptionally closely related with IoUAV. The combination of the IoT with ML and keen computing, will give and intelligent driving framework. Machine learning calculations within the self-driving empower IoUAV to predicts the conceivable changes within the encompassing driving environment and give distinctive errands i.e., protest location and recognizable proof, in expansion to prediction of another UAV’s localization and development [72]. Numerous ML calculations can be utilized to supply the specified errands. Relapse calculations give a localization plans to create a forecast and highlight determination models for self-driving UAVs. Clustering calculation give a way to modeling approaches such as centroid based and various leveled for shrewdly localization [73]. For cleverly choice making, choice lattice calculations will offer assistance to recognizing, analyzing, and rating the execution of connections between sets of values and data.

For self-driving UAVs approach, ML provides an intelligent decision making processes depends on the UAVs observations from different devices like, cameras, sensors, GPS and radars. Information collected from the UAVs devices help in learning process to extract an accurate driving decisions [74]. Intelligent decisions can have processed by using perception planning or end to end learning. Deep learning enables to perform the perception planning for purpose of various learning alterations and for non-learning components based. In end to end learning, deep learning enables to directly mapping sensors data to control operations. In addition, backpropagation algorithm for end to end learning enables to upgrade the performance for complex learning processes [75]. For example, profound learning calculations able to discover a course between UAV begin position and a wanted area, which speaks to way arranging. It’s able to consider all conceivable deterrents that are show within the encompassing environment, in expansion to discover out a direction with free of collision course.

A. Location Prediction

Computerization is a critical advantage of IoUAV organize. An unmanned aerial vehicle (UAV) contains discernment framework to be able protest discovery and forecast. In most IoUAV applications, the behavior of UAVs depends on tactile information, and capacity of classify the objects within the encompassing environment. These components offer assistance to create independent UAV applications by utilizing a proficient UAVs behavior forecast and choice making [76]. Brilliantly forecast will offer assistance to optimize the choice making to UAVs directions to maintain a strategic distance from any dangers. Self-driving and independent IoUAV are depend on the area expectation. The forecast requires data around the position of UAV, encompassing UAVs behaviors, in addition to geometry and activity of flying in 3D or 4D. Various models are created to predict UAVs behavior i.e. orientation, maneuver, and conscious interaction. [77]. These models are classified as the location prediction inputs to build features learning (see Fig. 11). In later a long time, analysts attempt to utilize the ML expectation strategies to optimize the accuracy of area forecast.

Fig. 11

Flowchart of IoUAV location Prediction based on ML.

ML employments recorded UAVs chronicled versatility designs to foresee another area expectation concurring to search direction designs. This technique is depending on the accessibility of sufficient authentic direction information [78]. To have pick up precise expectation, ML gives an effective strategy to induce ride the issue of enduring from the information inadequately and little historical direction, in expansion to effect of obscure energetic settings activity streams, climate. ML empowers to consolidate this relevant data into the UAV development expectation [79]. ML makes a difference to show properties of relevant data between trends and validates learning by coordinating neural order-state with long-term memory (LSTM) to predict the next region. The LSTM can effectively consolidate heterogeneous highlights by coordinated the direction factors to viably foresee the followed area.

In IoUAV, the trajectory prediction method is based LSTM network for tested on the flight trajectory recorded by the ground stations. Other methods can be used for such trajectory purpose such as Markov Model (MM). Figure 12 shows the performance of Markov and LSTM Models for IoUAV for flight trajectory prediction. Machine learning methods are used to predicts trajectory with the ground stations. The dynamic warping time helps to calculates the similarity between predicted trajectory and the round truth, enable to optimize the match between given sequences with certain restrictions.

Fig. 12

Performance Comparison Between MM and LSTM for IoUAVs Trajectory Prediction.

B. Resource Management

In IoUAV applications, asset administration is confronting numerous challenges because of dependence on IoT and large scale UAV network. asset administration challenges are related to gigantic channel get to, control assignment and obstructions administration, vitality administration, in addition to coexistence between the UAV and IoT. Gigantic channel getting to causes over-burdens to systems and blockage [80]. For asset administration, there’s a have to be create legitimate stack adjusting and get to administration procedures. The swarmed an unmanned aerial vehicle (UAV) traveling over the streets make impedances problems become which needs a proficient control assignment and impedances administration procedures. Within the IoUAV, the nature of the IoT is characterized by ceaseless information activity, leads to tall vitality utilization. Besides, the concordant coexistence between the UAVs existing systems and IoT, require a shrewdly asset administration [81].

ML calculations play a vital part in tending to the specified challenges related to asset management. For cleverly asset administration, ML able to form classification, relapse and thickness estimation to abuse the activity of information and create robotized arrangements for IoUAV administrations. ML gives brilliantly expectation for obscure IoUAV framework parameters framework behavior, i.e. support learning (RL) can empower to control the choices of framework activities from obscure checked framework behavior amid arrange exercises [82]. Besides, ML gives reasonable arrangements for empowering cautious channel and control allotment and deduce the UAV activity characteristics, and requests of the UAV’s clients. Profound learning promising keen arrangements to describe the distinct relationships between the inputs fo IoUAV framework and yield to create activity control framework in arrange to optimize the organize administration directing and planning adaption [83]. This will support to enhance the IoUAV network QoS.

Other thought related to the ML utilize in asset administration is to maximize the generally organize capacity and ensure best QoS. Q learning and deep reinforcement learning (DRL) can accomplish an incredible direction and technique by utilizing the arrange learning approach to achieve shrewd asset control, task and administration with continuous value exercises. Q-learning utilized to get an ideal model for asset allotment in UAV to UAV communications for anticipated amassed reduced rewards for long term. The Q work is approximated by a profound neural organize [84]. The ideal arrangement with Q-values can be found by the taking after condition.

$\begin{matrix} Q_{new} (s_{t}, a_{t}) = Q_{old} (s_{t}, a_{t}) + α [r_{t + 1} \\ + λ \max_{s ɛ S} Q_{old} (s, a_{t}) Q_{old} (s_{t}, a_{t})] \end{matrix}$ (3)

The watched state speaks to by sɛS, S speaks to the state space. t signified for time. s_t is a specialist state and a_t speaks to activity. The Q learning sent by actor-critic learning calculation (AC) [85]. The AC learning outline comprises of performing artist and pundit parts. These parts are dependable for choice the observation methodology with activity determinations, which is based on the order state tried and the policy entered for the environment parameter compensates for the action individually. This component empowers the IoUAV UAVs to form choices itself based on its learned arrangement technique [86].

DRL-based method in fixed base stations enables to solving the joint resource allocation problems. For clustering IoUAVs. Q-learning and DRL based clustering model to optimize the joint strategy for resource management. Figure 13 shows the performance example for Q-learning and DRL algorithms for a UAVs of two clustering groups. Q- learning can provide accuracy and precision up to 85%, however, DRL achieves more than 90% for accuracy, precision for UAVs clustering based network.

Fig. 13

Q-Learning and DRL Performances in IoUAVs Based Clustering Network.

Each IoUAV communication interface will watch the current arrange state i.e. asset square assignment, channel quality and the prerequisites of QoS to empower choice of activities related to asset square task and control level agreeing to the approach methodology and give unused IoUAV arrange state [87]. Due of the expanding mishaps and the critical have to be decrease street mischances and make strides activity security, present day an unmanned aerial vehicle (UAVs) is prepared with a sensors set, in addition to versatile communication systems. The UAV sensors allow the collection of a huge sum of information that’s utilized in UAV safety analysis methods. The information analyzed in genuine time by AI calculations within the independent driving frameworks to have high security levels by procedures to assign the secure monitoring records and forecast the parameter related to trajectory design, behavior of humans and activity stream. The depiction of trajectory security can be performed by the deep learning, which will foresee the real-time trajectory security list based on the neural arrange. In addition, For IoUAVs to IoV application, ML makes a difference in learning the link between visuals and city characteristics to measure street safety based on image handling [88]. Extracting the correlations between captured images and a street security assessment with many components across scales can achieve a high predictive accuracy of the Street Security (SI) record. Real-time estimation of street safety record will upgrade UAV security.

8 Machine learning based IoUAV for smart cities applications

Unmanned aerial vehicle (UAV), provide many applications for smart cities and create a positive impact on the society. The integration of UAVs in smart cities is very challenging due to several issues and concerns such as safety, privacy. IoUAV is used to deploy a quickly collect, aggregate, analyses and deliver highly accurate and highly detailed data. This data facilitates applications that improve operations, engage residents and support communities [89]. IoUAVs integrated with AI technologies can provide a best solutions and services related to areas, i.e. surveillance and security, inspection and detection, surveying and mapping, transport and delivery, farming and agriculture.

IoUAV networks can provides many smart cities applications, such as smart traffic management, natural disaster management, smart transportation, smart resource and asset management [90]. In smart traffic, UAVs enables to guide the people on the ground through from sky overview, and continuously and get a more comprehensive look at ground traffic of vehicles. In addition, IoUAVs provides flexible routes surveyors to map for any ground monitoring projects to take in-depth data to aid decision-making at an earlier stage. For disaster management, IoUAVs provides an intelligent monitoring model for the life threating situations and disaster, in addition to remote the information’s to the rescue’s centers with fast responses [91].

In IoUAV applications on smart cities, artificial intelligence (AI), machine learning (ML) and deep enhanced learning (DRL) play an important role in raising the design efficiency of UAVs systems and finding solutions to many complex problems directed towards the smart city [92]. The use of ML adds a great deal of intelligence to developing Intelligent Transportation Systems (ITS), and it also addresses a number of issues related to city security, in addition to city healthcare. Moreover, ML helps greatly in improving the power levels of unmanned aerial vehicles (UAVs) to ensure the best networks of 5 G and beyond.

In smart city applications, the process of data collection is one of the most important issue that achieve the characteristics of a smart city and meet the areas of application. When using UAVs, there must be a collaborative formula for data collection, storage and processing in real time to ensure that the required tasks can be carried out [107]. A number of IoT devices are used to perform these tasks, but they have a limited capacity relative to the energy of the batteries used on them, which makes them unable to transmit the signal over long distances, due to power limitations.

UAVs are usually connected to ground communication platforms, where they act as radio relays to improve the communication of ground radio devices and expand network coverage, or they can act as mobile air base stations to provide ground communications [107]. In both scenarios, energy consumption in UAVs will effects on the cost of energy-efficient when collects data from ground devices or when communicate with each other. Given that ground IoT devices can be mobile or static and provide data, UAVs must be optimally mobile to establish reliable and highly responsive communications to navigate the network. To reduce the power consumption of the UAVs in the mentioned scenarios by reducing the UAVs flight time, the concept of assigning the UAVs to a group of IoT ground devices can be used, so that one UAV is assigned to be the leader of each group when connected to other IoT devices as shown in Fig. 14.

Fig. 14

UAV network based on Clustered IoT Devices.

Due to the possibility of moving IoT devices according to smart cities environment nature, the composition of the IoT devices groups can change [108]. Therefore, it is necessary to update the locations of the UAVs for each corresponding IoT device group, while maintaining as much as possible the lowest energy consumption when changing the locations of the UAVs. Although the method of grouping IoT smart devices in clusters and communicating with the UAVs according to this concept, it is difficult to adopt this method in highly dynamic environments. This is because the current algorithms cannot perform the full processing and take advantage of the available data accurately [109]. Especially when considering the energy calculations according to the movement of UAVs and rearranging them according to the change in the IoT devices clustering environment, as well as the considerations of the communication channel calculations.

Recently, AI technologies such as ML have emerged as powerful methods that help in designing supervised and reinforced learning UAVs networks in different communication scenarios for smart city devices [110]. The use of ML enables to calculates the flight time for UAVs and adapts the UAVS speed according to the collecting and relaying the data process. Machine learning techniques help to design the UAVs trajectory and improve the process of collecting and transmitting data from multiple IoT devices and sensors that are widespread in smart cities. The RL algorithm can also be used to enable UAVs to adapt their paths to changes in the ground IoT devices, in addition to other changes related to channel status information and user requirements in terms of delay [111].

On the other hand, the ML algorithm enables to evaluate the transmission power of the UAVs, where the probability of interruption due to low power is known as a loss function that must be reduced according to the IoT devices locations, their density, the size of the coverage area, the communication requirements and the mutual distances between the UAVs. According to study in [112], ML can achieve better performance in reducing the power consumption in transmitting when compared with traditional power processing and management schemes. Figure 15 shown the performance of ML for transmitting power efficiency corresponding to the number of UAVs. As can be seen in the figure, machine learning maintains a low total power value required with low average communication power. Even when the number of UAVS are increased, they can serve a group of IoT devices in a small coverage area, which it helps to reduce energy consumption due to path loss and thus reduces the total transmission energy. In general, the use of ML will lead to a significant improvement in power consumption and reduce it to approximately 22%, and an even greater improvement can be given when the number of UAVs in the network increases. Increasing the deployment of more UAVs will reduce their movement over long distances due to their small coverage area, which can improve the UAVs power efficiency.

Fig. 15

Performance of ML Against Traditional Schemes for UAVS efficient transmit power.

9 Machine learning for IoUAV challenges and future trends

In IoT applications, the counterfeit insights play an incredible part that depend on discernment and expectations of occasions. In IoUAV applications, systems require an advancement of keen calculations to oversee cleverly innovation, such as in case of self-driving vehicle application. Self-driving approach is considered risky test for ML specialists, and for social learning cases in innovation administration [93]. In IoUAV, the integration of ML and the web of things guarantees future advance in productivity, precision and progressed asset administration. The utilize of ML with IoUAV provides tall execution within the field of communication and computing in arrange to attain productive control, administration and decision-making forms [94]. ML permits the extraction of enormous tactile information to induce way better bits of knowledge into the extend of issues related with the IoUAV and the encompassing environment, as well as the capacity to create basic operational choices. It moreover guarantees within the close future to overhaul the execution of UAV systems and enables interaction intuitively with Web based applications.

In expansion, the use of ML within IoUAV enables the interaction of cyber and physical components together, in addition to fundamentally improve the efficiency and consistent quality of models and frameworks [95]. Besides, ML provides shrewd arrangements to advance decision making in the occasion of cyber-attacks. Machine learning gives arrangements for numerous intelligent transportation applications, particularly in 2D perception of the level and determining. In any case, it can create AI methods that have the capacity to create applications enables collaborative portability depends on depiction of practical 3D and 4D objects discernments for driving independency [96]. For various IoUAV applications, such as driving administrations, and localization forecasting, the intelligent transportation system cameras devices can render visualizations to provide a 3D protest visualization. Given the half-breed setup for intelligent transport systems, the information mix is a distinct asset to move forward 3D visualization precision is an energizing conditional arrangement and basic future investigate heading. The 5 G of IoUAV organize is anticipated to supply some AI innovations to supply organize administration in a totally savvy and give imaginative administrations, but the 6th era (6 G) is anticipated to pack machine learning methods a vital part in its operation through self-reconfiguration on request to guarantee a multiplying in organize execution and benefit sorts [97]. ML procedures can give 6 G organize show that have the capacity to quickly react to IoUAV administration forms by learning in genuine time the state of the arrange.

10 Conclusion

ML is a foremost capable device for cleverly estimating and choice making that can help in analyzing enormous information in IoUAV systems. Numerous potential ML-based IoUAV applications have been tended to progress the execution of IoUAV systems. ML advances offer arrangements that are amazingly valuable in tending to blockage issue in tall thickness and fast topology alter IoUAV systems in arrange to guarantee quality of administrations and involvement. In addition, the scope of utilizing machine learning innovation in arrange administration and control, information stream, location estimating and asset instruments over diverse layers of communication systems were examined.

In general, we found, most computerized learning applications, implementation is dependent on the sums of information accessible where more information accessible way better execution would be achieved. As of late, ML and parallel computing capabilities have been created to construct savvy and coordinates IoUAV systems and frameworks. Different operations related to IoUAV such as multidimensional signals, remote communications, and image preparation are requiring broad computing and information preparing. vitality expending productivity one of the tricky issues that require ML solution.

Footnotes

Acknowledgments

This Project was funded by the Deanship of Scientific Research (DSR), King Abdulaziz University, Jeddah, under grant No. (G: 224-611-1441). The authors, therefore, gratefully acknowledge the DSR technical and financial support.

References

Yang

, Gao

, Zhang

, Yan

and He

, Codeword Selection for Concurrent Transmissions in UAV Networks: A Machine Learning Approach, IEEE Access 8 (2020), 26583–26590.

Khan

A.S.

, Chen

, Rahulamathavan

, Zheng

, Assadhan

and Lambotharan

, Trusted UAV Network Coverage Using Blockchain, Machine Learning, and Auction Mechanisms, IEEE Access 8 (2020), 118219–118234.

Chen

, Saad

and Yin

, Liquid State Machine Learning for Resource and Cache Management in LTE-U Unmanned Aerial Vehicle (UAV) Networks, IEEE Transactions on Wireless Communications 18(3) (2019), 1504–1517.

Alipour-Fanid

, Dabaghchian

, Wang

, Zhao

and Zeng

, Machine Learning-Based Delay-Aware UAV Detection and Operation Mode Identification Over Encrypted Wi-Fi Traffic, IEEE Transactions on Information Forensics and Security 15 (2020), 2346–2360.

Goudos

S.K.

and Athanasiadou

, Application of an Ensemble Method to UAV Power Modeling for Cellular Communications, IEEE Antennas and Wireless Propagation Letters 18(11) (2019), 2340–2344.

Wei

, Wu

, Ma

and Li

, State of charge prediction for UAVs based on support vector machine, The Journal of Engineering 23 (2019), 9133–9136.

Nguyen

V.N.

, Jenssen

and Roverso

, Intelligent Monitoring and Inspection of Power Line Components Powered by UAVs and Deep Learning, IEEE Power and Energy Technology Systems Journal 6(1) (2019), 11–21.

Lin

, Wang

, Zhou

, Ding

and Mao

, Dynamic Spectrum Interaction of UAV Flight Formation Communication with Priority: A Deep Reinforcement Learning Approach, IEEE Transactions on Cognitive Communications and Networking 6(3) (2020), 892–903.

, Zhang

, Wu

and Song

, Cellular UAV-to-Device Communications: Trajectory Design and Mode Selection by Multi-Agent Deep Reinforcement Learning, IEEE Transactions on Communications 68(7) (2020), 4175–4189.

10.

Borup

K.T.

, Fossen

T.I.

and Johansen

T.A.

, A Machine Learning Approach for Estimating Air Data Parameters of Small Fixed-Wing UAVs Using Distributed Pressure Sensors, IEEE Transactions on Aerospace and Electronic Systems 56(3) (2020), 2157–2173.

11.

Chen

, Saad

and Yin

, Echo-Liquid State Deep Learning for 360Â° Content Transmission and Caching in Wireless VR Networks with Cellular-Connected UAVs, IEEE Transactions on Communications 67(9) (2019), 6386–6400.

12.

Yin

, Wang

, Xu

, Liu

, Wang

and Wu

, Robust Visual Detection Learning, “Tracking Framework for Autonomous Aerial Refueling of UAVs”, IEEE Transactions on Instrumentation and Measurement 65(3) (2016), 510–521.

13.

Liu

C.H.

, Chen

, Tang

, Xu

and Piao

, Energy-Efficient UAV Control for Effective and Fair Communication Coverage: A Deep Reinforcement Learning Approach, IEEE Journal on Selected Areas in Communications 36(9) (2018), 2059–2070.

14.

Zhang

, Yu

, Mao

, Periaswamy

S.C.G.

, Patton

and Xia

, IADRL: Imitation Augmented Deep Reinforcement Learning Enabled UGV-UAV Coalition for Tasking in Complex Environments, IEEE Access 8 (2020), 102335–102347.

15.

Tanwar

, Bhatia

, Patel

, Kumari

, Singh

P.K.

and Hong

, Machine Learning Adoption in Blockchain-Based Smart Applications: The Challenges, and a Way Forward, IEEE Access 8 (2020), 474–488.

16.

and Li

, Research on a Single-Tree Point Cloud Segmentation Method Based on UAV Tilt Photography and Deep Learning Algorithm, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 13 (2020), 4111–4120.

17.

Luo

, Zhang

, He

, Yang

and Ji

, A Two-Step Environment-Learning-Based Method for Optimal UAV Deployment, IEEE Access 7 (2019), 149328–149340.

18.

Liu

, Liu

, Chen

and Hanzo

, Trajectory Design and Power Control for Multi-UAV Assisted Wireless Networks: A Machine Learning Approach, IEEE Transactions on Vehicular Technology 68(8) (2019), 7957–7969.

19.

Malek

, Bazi

, Alajlan

, AlHichri

and Melgani

, Efficient Framework for Palm Tree Detection in UAV Images, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 7(12) (2014), 4692–4703.

20.

Shan

, Miura

, Kagawa

, Ono

, Li

and Kojima

, Machine Learning-Based Field Data Analysis and Modeling for Drone Communications, IEEE Access 7 (2019), 79127–79135.

21.

Bazi

and Melgani

, Convolutional SVM Networks for Object Detection in UAV Imagery, IEEE Transactions on Geoscience and Remote Sensing 56(6) (2018), 3107–3118.

22.

Zeng

, Wang

and Xu

, A Basal Ganglia Network Centric Reinforcement Learning Model and Its Application in Unmanned Aerial Vehicle, IEEE Transactions on Cognitive and Developmental Systems 10(2) (2018), 290–303.

23.

Hoang

T.M.

, Nguyen

N.M.

and Duong

T.Q.

, Detection of Eavesdropping Attack in UAV-Aided Wireless Systems: Unsupervised Learning with One-Class SVM and K-Means Clustering, IEEE Wireless Communications Letters 9(2) (2020), 139–142.

24.

Zhu

, Gui

, Cheng

, Sun

and Zhang

, Joint Design of Access Point Selection and Path Planning for UAV-Assisted Cellular Networks, IEEE Internet of Things Journal 7(1) (2020), 220–233.

25.

Yang

, Zhang

, He

, Wen

, Ji

and Li

, Machine-learning-based prediction methods for path loss and delay spread in air-to-ground millimetre-wave channels, IET Microwaves, Antennas & Propagation 13(8) (2019), 1113–1121.

26.

Rajagopal

, Joshi

G.P.

, Ramachandran

, Subhalakshmi

R.T.

, Khari

, Jha

, Shankar

and You

, A Deep Learning Model Based on Multi-Objective Particle Swarm Optimization for Scene Classification in Unmanned Aerial Vehicles, IEEE Access 8 (2020), 135383–135393.

27.

Mozaffari

, Saad

, Bennis

, Nam

and Debbah

, A Tutorial on UAVs for Wireless Networks: Applications, Challenges, and Open Problems, IEEE Communications Surveys & Tutorials 21(3) (2019), 2334–2360.

28.

Anwar

M.Z.

, Kaleem

and Jamalipour

, Machine Learning Inspired Sound-Based Amateur Drone Detection for Public Safety Applications, IEEE Transactions on Vehicular Technology 68(3) (2019), 2526–2534.

29.

, Ren

, Cheng

, Xue

, Chen

, Debbah

and Han

, Millimeter-Wave Networking in the Sky: A Machine Learning and Mean Field Game Approach for Joint Beamforming and Beam-Steering, IEEE Transactions on Wireless Communications 19(10) (2020), 6393–6408.

30.

Ezuma

, Erden

and Kumar

, Anjinappa, O. Ozdemir and I. Guvenc, Detection and Classification of UAVs Using RF Fingerprints in the Presence of Wi-Fi and Bluetooth Interference, IEEE Open Journal of the Communications Society 1 (2020), 60–76.

31.

Zhao

, Liu

and Cheng

, Multi-Agent Deep Reinforcement Learning for Trajectory Design and Power Allocation in Multi-UAV Networks, IEEE Access 8 (2020), 139670–139679.

32.

Koda

, Zeggada

, Melgani

and Nishii

, Spatial and Structured SVM for Multilabel Image Classification, IEEE Transactions on Geoscience and Remote Sensing 56(10) (2018), 5948–5960.

33.

Zheng

, Du

, Ling

, Sheng

and Chen

, Evolutionary Collaborative Human-UAV Search for Escaped Criminals, IEEE Transactions on Evolutionary Computation 24(2) (2020), 217–231.

34.

Cheng

, Lyu

, Quan

, Zhou

, He

, Shi

and Shen

, Space/Aerial-Assisted Computing Offloading for IoT Applications: A Learning-Based Approach, IEEE Journal on Selected Areas in Communications 37(5) (2019), 1117–1129.

35.

Huang

, Xu

, Zhang

, Hua

and Zhang

, An Overview of Intelligent Wireless Communications using Deep Reinforcement Learning, Journal of Communications and Information Networks 4(2) (2019), 15–29.

36.

Rajagopal

, Ramachandran

, Shankar

, Khari

, Jha

, Lee

and Joshi

G.P.

, Fine-Tuned Residual Network-Based Features with Latent Variable Support Vector Machine-Based Optimal Scene Classification Model for Unmanned Aerial Vehicles, IEEE Access 8 (2020), 118396–118404.

37.

Shiri

, Park

and Bennis

, Remote UAV Online Path Planning via Neural Network-Based Opportunistic Control, IEEE Wireless Communications Letters 9(6) (2020), 861–865.

38.

James

, Ford

J.J.

and Molloy

T.L.

, Learning to Detect Aircraft for Long-Range Vision-Based Sense-and-Avoid Systems, IEEE Robotics and Automation Letters 3(4) (2018), 4383–4390.

39.

Liu

, Chen

and Yin

, Optimized Trajectory Design in UAV Based Cellular Networks for 3D Users: A Double Q-Learning Approach, Journal of Communications and Information Networks 4(1) (2019), 24–32.

40.

Pham

Q.-V.

, Huynh-The

, Alazab

, Zhao

and Hwang

W.-J.

, Sum-Rate Maximization for UAV-Assisted Visible Light Communications Using NOMA: Swarm Intelligence Meets Machine Learning, IEEE Internet of Things Journal 7(10) (2020), 10375–10387.

41.

Pan

, Zhang

, Cervone

and Yang

, Detection of Asphalt Pavement Potholes and Cracks Based on the Unmanned Aerial Vehicle Multispectral Imagery, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 11(10) (2018), 3701–3712.

42.

Zhang

, Chen

, Wang

and Dai

, Object-Based Land-Cover Supervised Classification for Very-High-Resolution UAV Images Using Stacked Denoising Autoencoders, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 10(7) (2017), 3373–3385.

43.

Egi

and Otero

C.E.

, Machine-Learning and 3D Point-Cloud Based Signal Power Path Loss Model for the Deployment of Wireless Communication Systems, IEEE Access 7 (2019), 42507–42517.

44.

Shakeri

, Al-Garadi

M.A.

, Badawy

, Mohamed

, Khattab

, Al-Ali

A.K.

, Harras

K.A.

and Guizani

, Design Challenges of Multi-UAV Systems in Cyber-Physical Applications: A Comprehensive Survey and Future Directions, IEEE Communications Surveys & Tutorials 21(4) (2019), 3340–3385.

45.

Keneni

B.M.

, Kaur

, Al Bataineh

, Devabhaktuni

V.K.

, Javaid

A.Y.

, Zaientz

J.D.

and Marinier

R.P.

, Evolving Rule-Based Explainable Artificial Intelligence for Unmanned Aerial Vehicles IEEE Access 7 (2019), 17001–17016.

46.

Valente

, Kooistra

and Mcher

, Fast Classification of Large Germinated Fields Via High-Resolution UAV Imagery, IEEE Robotics and Automation Letters 4(4) (2019), 3216–3223.

47.

Bah

M.D.

, Hafiane

and Canals

, CRowNet: Deep Network for Crop Row Detection in UAV Images, IEEE Access 8 (2020), 5189–5200.

48.

Luo

, Yu

and Ren

, A Vision-Aided Approach to Perching a Bioinspired Unmanned Aerial Vehicle, IEEE Transactions on Industrial Electronics 65(5) (2018), 3976–3984.

49.

Chen

, Mozaffari

, Saad

, Yin

, Debbah

and Hong

C.S.

, Caching in the Sky: Proactive Deployment of Cache-Enabled Unmanned Aerial Vehicles for Optimized Quality-of-Experience, IEEE Journal on Selected Areas in Communications 35(5) (2017), 1046–1061.

50.

Wang

, Yang

, Liu

and Gui

, LightAMC: Lightweight Automatic Modulation Classification via Deep Learning and Compressive Sensing, IEEE Transactions on Vehicular Technology 69(3) (2020), 3491–3495.

51.

Shang

, Liu

, Rao

R.M.

, Marojevic

and Reed

J.H.

, 3D Spectrum Sharing for Hybrid D2D and UAV Networks, IEEE Transactions on Communications 68(9) (2020), 5375–5389.

52.

Wei

, Yang

, Cao

, Lu

and Wang

, Recurrent MADDPG for Object Detection and Assignment in Combat Tasks, IEEE Access 8 (2020), 163334–163343.

53.

Mozaffari

, Taleb Zadeh Kasgari

, Saad

, Bennis

and Debbah

, Beyond 5G With UAVs: Foundations of a 3D Wireless Cellular Network, IEEE Transactions on Wireless Communications 18(1) (2019), 357–372.

54.

, Melgani

and He

, CSVM Architectures for Pixel-Wise Object Detection in High-Resolution Remote Sensing Images, IEEE Transactions on Geoscience and Remote Sensing 58(9) (2020), 6059–6070.

55.

Gao

, Shen

and Niu

, Large-Scale Synthetic Urban Dataset for Aerial Scene Understanding, IEEE Access 8 (2020), 42131–42140.

56.

Barmpounakis

E.N.

, Vlahogianni

E.I.

and Golias

J.C.

, Identifying Predictable Patterns in the Unconventional Overtaking Decisions of PTW for Cooperative ITS, IEEE Transactions on Intelligent Vehicles 3(1) (2018), 102–111.

57.

PaweÅ,czyk

and Wojtyra

, Real World Object Detection Dataset for Quadcopter Unmanned Aerial Vehicle Detection, IEEE Access 8 (2020), 174394–174409.

58.

Martinez-Cantin

, Funnelled Bayesian Optimization for Design, Tuning and Control of Autonomous Systems, IEEE Transactions on Cybernetics 49(4) (2019), 1489–1500.

59.

Chen

C.-J.

, Huang

Y.-Y.

, Li

Y.-S.

, Chang

C.-Y.

and Huang

Y.-M.

, An AIoT Based Smart Agricultural System for Pests Detection, IEEE Access 8 (2020), 180750–180761.

60.

Chang

, You

and Cao

, Bidimensional Empirical Mode Decomposition for SAR Image Feature Extraction with Application to Target Recognition, IEEE Access 7 (2019), 135720–135731.

61.

, Xu

, Zhang

, Yin

, Chen

and Han

, A Prediction-Based Charging Policy and Interference Mitigation Approach in the Wireless Powered Internet of Things, IEEE Journal on Selected Areas in Communications 37(2) (2019), 439–451.

62.

Shoufan

, Al-Angari

H.M.

and Sheikh

M.F.A.

, E., Damiani Drone Pilot Identification by Classifying Radio-Control Signals, IEEE Transactions on Information Forensics and Security 10 (2018), 2439–2447.

63.

, Han

, Tao

, Zhang

, Grecos

, Plaza

and Ren

, Hashing Nets for Hashing: A Quantized Deep Learning to Hash Framework for Remote Sensing Image Retrieval, IEEE Transactions on Geoscience and Remote Sensing 58(10) (2020), 7331–7345.

64.

, Xin

, Chen

, Xin

, Wei

, Li

, Zhang

, Jin

, Tu

, Zhou

and Liao

, An Automatic Detection Method of Bird, “Nest on Transmission Line Tower Based on Faster RCNN”, IEEE Access 8 (2020), 164214–164221.

65.

Cao

, Chang

and Raychowdhury

, A 65-nm 8-to-3-b 1.0 0.36-V 9.1 1.1-TOPS/W Hybrid-Digital-Mixed-Signal Computing Platform for Accelerating Swarm Robotics, IEEE Journal of Solid-State Circuits 55(1) (2020), 49–59.

66.

Rohan

, Rabah

and Kim

, Convolutional Neural Network-Based Real-Time Object Detection and Tracking for Parrot AR Drone 2, IEEE Access 7 (2019), 69575–69584.

67.

Baraković

and Skorin-Kapov

, Survey and Challenges of QoE Management Issues in Wireless Networks, Journal of Computer Networks and Communications 2013, Article ID 165146, (2013), 28. https://doi.org/10.1155/2013/165146.

68.

Bhardwaj

C.G.A.

, Applications of Unmanned Aerial Vehicles in Cryosphere: Latest Advances and Prospects, Remote Sens 12 (2020), 948. doi: 10.3390/rs12060948

69.

García

, et al., Dron Away: A Proposal on the Use of Remote Sensing Drones as Mobile Gateway for WSN in Precision Agriculture, Appl Sci 10 (2020), 6668. doi: 10.3390/app10196668

70.

Koubaa

, Ammar

, Alahdab

, Kanhouch

and Azar

A.T.

, DeepBrain:ExperimentalEvaluation of Cloud-Based Computation Offloading and Edge Computing in the Internet-of-Drones for Deep Learning Applications, Sensors (Basel) 20(18) (2020), 5240. Published 2020 Sep 14. doi:10.3390/s20185240

71.

Outay

, Mengash

H.A.

and Adnan

, Applications of unmanned aerial vehicle (UAV) in road safety, traffic and highway infrastructure management: Recent advances and challenges, Transp Res Part A Policy Pract 141 (2020), 116–129. doi: 10.1016/j.tra.2020.09.018

72.

Zhang

and Letaief

K.B.

, Mobile Edge Intelligence and Computing for the Internet of Vehicles, rXiv:1906.00400v1, (2019).

73.

Zargari

S.A.

and Safari

, Using Clustering Methods in Multinomial Logit Model for Departure Time Choice, Journal of Advanced Transportation 2020, Article ID 7382569, 12 pages, 2020. https://doi.org/10.1155/2020/7382569.

74.

Grigorescu

, et al., A Survey of Deep Learning Techniques for Autonomous Driving, arXiv:1910.07738v2, (2020).

75.

Liu

, et al., Unmanned Aerial Vehicle for Internet of Everything: Opportunities and Challenges, arXiv:2003.13311v2, (2020).

76.

Hazim , et al., Unmanned Aerial Vehicles (UAVs): A Survey on Civil Applications and Key Research Challenges, IEEE access (2019).

77.

Fraga-Lamas

, Ramos

, Mondéjar-Guerra

and Fernández-Caramés

T.M.

, A Review on IoT Deep Learning UAV Systems for Autonomous Obstacle Detection and Collision Avoidance, Remote Sens 11 (2019), 2144. https://doi.org/10.3390/rs11182144.

78.

Sandino

, et al., UAV Framework for Autonomous Onboard Navigation and People/Object Detection in Cluttered Indoor Environments, Remote Sens 12 (2020), 3386. doi: 10.3390/rs12203386

79.

W.S.

, et al., Impedance-Based Non-Destructive Testing Method Combined with Unmanned Aerial Vehicle for Structural Health Monitoring of Civil Infrastructures, Appl Sci 7 (2017), 15. doi: 10.3390/app7010015

80.

Kiela

, et al., Review of V2X–IoT Standards and Frameworks forITS Applications, Appl Sci 10 (2020), 4314. doi: 10.3390/app10124314

81.

Bithas

P.S.

, Michailidis

E.T.

, Nomikos

, Vouyioukas

and Kanatas

A.G.

, A Survey on Machine-Learning Techniques for UAV-Based Communications,Published Nov, Sensors (Basel) 19(23) (2019), 5170. Published 2019 Nov 26. doi:10.3390/s19235170

82.

Abdalla

l.S.

and Marojevic

, Machine Learning-Assisted UAV Operations with UTM: Requirements, Challenges, and Solutions, arXiv:2006.14544v1, (2020).

83.

Bucknell

and Bassindale

, An investigation into the effect of surveillance drones on textile evidence at crime scenes, Sci Justice 57(5) (2017), 373–375. doi: 10.1016/j.scijus.2017.05.004. Epub 2017 May 15.

84.

, Wang

, Mo

, et al., Robust proactive eavesdropping in UAV-enabled wireless communication networking, J Wireless Com Network 2019 (2019), 285. https://doi.org/10.1186/s13638-019-1599-6.

85.

Fatemi

, et al., Policy Networks with Two-Stage Training for Dialogue Systems, Proceedings of the SIGDIAL 2016 Conference 101–110.

86.

, Ren

and Dong

, Supervised-actor-critic reinforcement learning for intelligent mechanical ventilation and sedative dosing in intensive care units, BMC Med Inform Decis Mak 20 (2020), 124. https://doi.org/10.1186/s12911-020-1120-5.

87.

Chuang

H.-M.

, He

and Namiki

, Utonomous Target Tracking of UAV Using High-Speed Visual Feedback, Appl Sci 9 (2019), 4552. doi: 10.3390/app9214552

88.

Ding

, et al., Frontier Progress of Unmanned Aerial Vehicles Optical Wireless Technologies, Sensors 20 (2020), 5476. doi: 10.3390/s20195476

89.

Mohamed

, Al-Jaroodi

, Jawhar

, Idries

and Mohammed

, Unmanned aerial vehicles applications in future smart cities, Technological Forecasting and Social Change 153 (2020).

90.

Impedovo

and Pirlo

, Artificial Intelligence Applications to Smart City and Smart Enterprise, Appl Sci 10 (2020), 2944. doi: 10.3390/app10082944

91.

, Qurishee

M.A.

, Owino

, Fomunung

, Onyango

and Atolagbe

, Coupling Deep Learning and UAV for Infrastructure Condition Assessment Automation, 2018 IEEE International Smart Cities Conference (ISC2), Kansas City, MO, USA, (2018), 1–7, doi: 10.1109/ISC2.2018.8656971

92.

Awan

, et al., Machine Learning-Enabled Power Scheduling inIoT-Based Smart Cities, Computers, Materials & Continua, (2021).

93.

Jing

, et al., He Determinants behind the Acceptance of Autonomous Vehicles: A Systematic Review, Sustainability 12 (2020), 492. doi: 10.3390/su12020492

94.

Davenport

and Kalakota

, The potential for artificial intelligence in healthcare, Future Healthcare Journal 6(2) (2019), 94–98. doi: 10.7861/futurehosp.6-2-94

95.

Tohidi

and Rustamov

R.B.

, A Review of the Machine Learning in GIS for Megacities Application, Geographic Information Systems in Geospatial Intelligence, Rustam B. Rustamov, IntechOpen, (2020), DOI: 10.5772/intechopen.94033. Available from: https://www.intechopen.com/books/geographic-information-systems-in-geospatial-intelligence/a-review-of-the-machine-learning-in-gis-for-megacities-application.

96.

Çinar

Z.M.

, et al., Machine Learning in Predictive Maintenance towards Sustainable Smart Manufacturing in Industry 4.0, Sustainability 12 (2020), 8211. doi: 10.3390/su12198211

97.

Ahuja

A.S.

, The impact of artificial intelligence in medicine on the future role of the physician, Peer J 7 (2019), e7702. doi: 10.7717/peerj.7702

98.

Khan

A.I.

and Al-Mulla

, Unmanned Aerial Vehicle in the Machine Learning Environment, Procedia Computer Science 160 (2019), 46–53.

99.

Jingzh , et al., Reinforcement Learning for a Cellular Internet of UAVs: Protocol Design, Trajectory Control, and Resource Management, arXiv:1911.08771v1, (2019).

100.

Ursula , et al., Machine Learning for Wireless Connectivity and Security of Cellular-Connected UAVs, arXiv:1804.05348v3, (2018).

101.

Kemp

N.A.

, Artificial Intelligence Empowered UAVs Data Offloading in Mobile Edge Computing, (2019). https://digitalrepository.unm.edu/ece_etds/479.

102.

Lee

H.Y.

, Ho

H.W.

and Zhou

, Deep Learning-based Monocular Obstacle Avoidance for Unmanned Aerial Vehicle Navigation in Tree Plantations, J Intell Robot Syst 101 (2021), 5 https://doi.org/10.1007/s10846-020-01284-z.

103.

Tang

, et al., Minimum Throughput Maximization for Multi-UAV Enabled WPCN: A Deep Reinforcement Learning Method, Special section on artificial intelligence for physical layer wireless communications, IEEE Access (2019).

104.

Bijolin Edwin

, Roshni Thanka

and Deula

, n Internet of Drone (IOD) Based Data Analytics in Cloud for Emergency Services, International Journal of Recent Technology and Engineering (IJRTE) ISSN: 7(5S2) (2019), 2277–3878.

105.

Meng

, Shafik

, Mojtaba Matinkhah

and Ahmad

, A 5G Beam Selection Machine Learning Algorithm for Unmanned Aerial Vehicle Applications, Wireless Communications and Mobile Computing 2020, Article ID 1428968, (2020), 16. https://doi.org/10.1155/2020/1428968.

106.

Ahlawat

, et al., Smart surveillance using on cloud machine learning and internet controlled UAVs, Advances in Mathematics Scientific Journal 6 (2020), 3789–3794.

107.

Alsamhi

S.H.

, Ma

, Ansari

M.S.

and Almalki

F.A.

, Survey on Collaborative Smart Drones and Internet of Things for Improving Smartness of Smart Cities, in , IEEE Access 7 (2019), 128125–128152. doi: 10.1109/ACCESS.2019.2934998

108.

Mozaffari

, Saad

, Bennis

and Debbah

, Mobile Internet of Things: Can UAVs Provide an Energy-Efficient Mobile Architecture? arXiv:1607.02766v1, (2016).

109.

Lagkas

, Argyriou

, Bibi

and Sarigiannidis

, UAV IoT Framework Views and Challenges: Towards Protecting Drones as “Things”, Sensors 18 (2018), 4015.

110.

Azar

A.T.

, Koubaa

, Ali Mohamed

, Ibrahim

H.A.

, Ibrahim

Z.F.

, Kazim

, Ammar

, Benjdira

, Khamis

A.M.

, Hameed

I.A.

and Casalino

and Ali Mohamed

, Drone Deep Reinforcement Learning: A Review, Electronics 10 (2021), 999.

111.

Kouhdaragh

, Verde

, Gelli

and Abouei

, On the Application of Machine Learning to the Design of UAV-Based 5G Radio Access Networks, Electronics 9 (2020), 689.

112.

Zhang

, et al., Machine Learning for Predictive On-Demand Deployment of UAVs for Wireless Communications, arXiv:1805.00061v1, (2018).