Machine Learning from Drone Forensics

Machine Learning from Drone Forensics: History

View Latest Version

Please note this is an old version of this entry, which may differ significantly from the current revision.

Contributor:

Zubair Baig

Nazeeruddin Mohammad

Ghassen Ben Brahim

drones
criminal activity
machine learning

1. Introduction

Drones are flying Internet of Things (IoT) objects that are an embodiment of hardware designed to be driven by software-based controls. Drones are programmed to fly according to user-defined specifications and have on-device IoT sensors and cameras augmented with a Global Positioning System (GPS) controller to facilitate their flight and all activity relevant to their operations. Drones are useful for several civilian applications. Some of the earlier work from the 1990s in unmanned aerial vehicle (UAV) technology ^[1] was specified for military applications alone. This comprised reconnaissance work for battlefields to enable military expeditions. However, contemporary drone utility in the civilian domain is gaining increasing acceptance by consumers.

Drone sales have only exponentially increased in the last five years, with sales topping eight million in 2021. According to ^[2], retail applications adopting drones will exceed 122,000 by 2023, with popular purposes being aerial photography, express delivery services, reconnaissance for disaster-hit zones, thermal sensing for rescue operations, building safety inspections, crop monitoring, storm tracking and border surveillance (law enforcement). Moreover, we live in an era of Generation 7 drones, which have the capability of intelligent operation, hazard avoidance, holistic airspace awareness and automated flight. Data that consequently are produced through drone flights are significant in volume.

Drones have been deployed to reach locations too hazardous for human access, such as facility observation at an altitude, surveillance and monitoring of high-rise buildings, analysis of mobile communication towers for anomalies and overhead electricity transmission line monitoring.

Smartphones play an integral role in the process of controlling drones. They serve dual purposes in the phone-to-drone interaction, where users can switch between manual and automatic/autonomous control modes. In ^[3], the studied drones sent light commands, drone status, images, and video over WIFI communication channels. Under this setup, a client smartphone may issue a set of predefined commands that vary the drone’s rotors to change the drone’s position while operating in the manual mode. Alternatively, image processing and machine-learning algorithms can run on the client smartphone, generating commands that return the drone to autonomous flight modes. For longer distance transmission of data between drones and smartphones, a different method of operation is employed. It relies on a 2.4 GHz radio communication channel between a transmitter and a receiver controller attached to a smartphone via a USB cable and a receiver mounted on the drone’s assembly ^[4]. Autonomous navigation of the drone can be achieved by having the smartphone gain access to the drone’s controls to issue commands based on flight calculations as a result of trajectory calculations and vision-based processing that run on the smartphone. The authors in ^[5] suggest building on top of these functions to deploy an autonomous landing system for drones.

Drones can breach airspace regulations of their jurisdictions as part of malicious attacks that can be perpetrated by a criminal, where the rogue agent can use a fake email address to log in to the mobile smart app of a drone and conceal its identity when it is carrying out some criminal action such as ‘breach of airspace’ or carrying out illegal activities such as taking photos of strategic or sensitive locales ^[6]. This threat is posed to vulnerable drones that have few or no security controls in place to prevent device compromise.

In the event of a drone being involved in criminal activities, its confiscation and subsequent analysis at a digital forensic investigation laboratory is a crucial part of evidence gathering and analysis. Such activity precedes any presentation of admissible evidence against the owner of a confiscated drone.

According to ^[7], challenges associated with drone forensics include:

Post-crash scattering of individual drone components encumbers routine association of parts to a drone seized at crash site.
The diverse types of on-device components for a drone imply that the use of a single digital forensic investigation tool will not serve the purpose of investigation; a full range of tools, both hardware and software, would be needed to run a thorough forensic procedure.
Physical data acquisition of forensic images from a drone may not be practicable as certain drones only permit wireless transfer of images.
Access control and protection mechanisms may prevent certain data elements from being acquired as part of the forensic image. Moreover, drone controller chips may be accessible only through an owner-signed remote controller, which can be difficult to emulate by law enforcement.;
Certain drones have multiple file systems on them, thus encumbering the process of identifying the right tool to be able to carry out data acquisition.
Add-on software makes it difficult to forecast the software platform, file system and the corresponding hardware configuration for a seized drone.
Flash memory and RAM can lose data after a crash, if the battery of the drone runs out;
Data logs may be partial or programmed to not hold any data depending on the drone model.
Deliberate attempts by a remote controller to wipe out data on a confiscated drone does not help the law enforcement procedure.

2. Machine Learning from Drone Forensics

2.1. Machine Learning Primer

Machine learning (ML) is a branch of artificial Intelligence that primarily focuses on making predictions (or forecasting) through the development of mathematical models. These models are designed in such a way that they explore abundant and massive amounts of data and attempt to exploit the inherent correlations within the various components of the data to identify repeating patterns. This helps with the process of decision making with little or no human intervention, i.e., automated decision making is made tangible. These models also try to learn from “experience” (also known as historic data) to improve prediction accuracy. Machine-learning algorithms comprise two parts: a training phase and a testing phase.

The process of improving prediction performance is carried out during the training phase of machine learning, where the algorithm is introduced with a large set of historic data typically in an iterative manner for producing mathematical values, to emulate an artificially trained brain. Applications of machine learning include speech recognition ^[8], natural language processing ^[9], robotic vehicles ^[10], fraud detection ^[11], text and handwriting classification ^[12], object classification ^[13], digital forensics ^[14] and systems security ^[15].

Machine-learning algorithms can also help uncover and learn hidden patterns embedded in the data under analysis or perform classification of observed data. This is the test phase of the process. Each of these algorithms implement a different philosophy on how data are analyzed. Example of such algorithms include decision trees, support vector machines, artificial neural networks, linear regression, K-nearest neighbor, naïve Bayes and random forest.

2.2. Machine Learning for Drone Data

Machine learning has been previously proposed to analyze several problem domains related to unmanned aerial vehicles (UAV). A detailed survey on ML techniques used for UAV-based communications is presented in ^[16]. The paper highlights how ML has been used for improving several communication concerns including channel modeling, resource management and positioning and security within UAV based communication. The paper classifies ML applications within four broader categories including: (1) physical layer aspects (channel modeling, interference management and spectrum allocation), (2) resource management aspects (such as network planning, power management, routing and data caching), (3) positioning (such as placement, detection and mobility), and (4) security (public safety, network jamming and eavesdropping). The paper then provides a summary of relevant work within each problem domain of these broader categories.

A jamming attack comprises an adversarial attempt to inject noise into a communication channel to cause disruption of routine communication exchange. In ^[17], a two classifier-based approach is proposed for detecting jamming attacks on a C-RAN network. The first classifer is a multilayer perceptron (MLP) and the second is a Kernlab support vector machine (KSVM). Jamming attacks were attested as not being linearly separable in a low dimension space. Therefore, the distinction between two classes of radio signal data is realizable through the adoption of a KSVM machine-learning solution for those jamming attack vectors that circumvent the MLP classifier. Results show promise and help prove the significance of adopting machine learning for classification of data to refer to a jamming or an eavesdropping attack.

In ^[18], an anomaly detection scheme is proposed for mitigating the effects of several attack vectors. The machine- learning-based anomaly detector is able to identify five attack types, namely constant position deviation attack (message modification), random position deviation attack (message modification), velocity drift attack (message modification), DOS attack (message deletion) comprising constructive and destructive interference, and the flight replacement attack (message injection). The use case analyzed is the air traffic surveillance system, ADS-B (automatic dependent surveillance-broadcast). The two-step anomaly detection scheme comprises preliminary reconstruction of ADS-B data, combined presentation of the reconstructed and the actual values to the SVDD (support vector data description) for training, and the definition and implmenetation of a hypersphere classifier for anomaly detection.

Reinforced learning-based power provision approaches are used to protect UAV transmissions against attacks such as eavesdropping and jamming ^[19]. ML can also be used for detecting an eavesdropper by building a classifier based on the received signals associated with eavesdropping attacks and non-attacks ^[20]. This activity is based upon prior training of ML models through presentation of data that depict a radio signal jamming attack to the ML classifier.

Another survey paper ^[21] focused on deep-learning techniques used in UAV problem domains for feature extraction, planning and situational awareness. In ^[22], the authors first highlighted that drones typically fly at an altitude that is higher than traditional ground user equipment. Radio signal propagation is affected through flight through height and also line of sight of free space propagation. A scheme is proposed for the identification of rogue drones that may be found in a mobile network. Legitimate drones may be registered with ground equipment. However, unregistered rogue drones permeating the airspace in sensitive locales may prove to be a security risk. The authors emulated drone deployment scenarios comprising outdoor drones and ground user equipment for urban scenarios. The simulation setup included the following parameters: number of flying sites and sectors, inter-site distance, antennas for a base station (height, power) and carrier frequencies. Measurement data were collected from the simulations and split into a training and a testing set. Two machine-learning techniques were adopted, namely logistic regression (LR) and decision trees (DT). For LR, two categories (variables) were specified, drones and other user equipment, respectively. DT are supervised-learning models that work on feature-value tuples extracted from a dataset. In this case, four features were observed, namely received signal strength indicator (RSSI), standard deviation of the eight strongest reference signals, difference between top two strength reference signals and serving cell values. Classification results yielded a 100% accuracy in detection of rogue drones for >60 m altitudes, and 5% detection rate for lower altitudes. This was attributed to the radio frequency interference phenomenon, which is more significant at lower altitudes.

In ^[23], a deep-learning-based approach is presented for drone detection and identification. In particular, drone acoustic fingerprints were analyzed for detection and identification. Specifications on drone noise data comprised foot printing of drones to produce 1300 audio clips of drone sounds. Furthermore, to ascertain accuracy in detection, the datasets thus derived were an amalgamation of pure drone noise, silence and drone audio clips that were captured through drone propeller noise generated in an indoor setting. Audio clips were also balanced based on time intervals between captures. Each audio file was processed based on file type, data sampling rate and the bitrate of the channel. Additionally, audio files were also segmented into smaller chunks (which were further experimented on to deduce the most accurate segment size) to improve the performance of the deep-learning classifier. Classification of the processed drone data by the three adopted classifiers, namely recurrent neural networks (RNN), convolutional neural network (CNN) and convolutional recurrent neural network (CRNN), were subsequently reported by the contributors when these were tested on a three-class classification experiment (drone type one, drone type two and other noise). Results portrayed the superiority of the CNN technique over the other two.

Lee et al. provide a comprehensive drone detection system using machine learning in ^[24]. The authors were able to classify camera-equipped drone data, i.e., image data, through the adoption of a cascade classification of images using CNNs. Drone data were manually labeled, comprising 2099 drone images. A total of 1777 were used for training and the remainder 429 for testing. The system was able to deduce the location of a drone on a camera-captured image as well as the vendor model of a drone based on machine classification with reported accuracies of >90%. For feature extraction, the authors were able to adopt the Haar feature processing method to extract drone sub-images from the image dataset obtained from ^[25].

In ^[26], an approach for identifying anomalies in a swarm flight comprising multiple flying drones, wherein certain drones may be deliberately controlled by the adversary to cause a possible sabotage, was proposed. Flight data from multiple streams were analyzed to identify such anomalies. Drone data comprising time-series sensory data are sampled at a certain frequency, with the authors generating 16 samples per time stamp. Data from normal and anomalous drones are prelabeled. Categories of anomalies were defined into three, namely noise caused through sensor generated signal disruption in flight, abnormal signals generated in actual flight but recoverable in flight and signal errors causing the aircraft to halt flight due to malfunction. The classifier selected for the experiments was the 1D signal unsupervised CNN based on a generative model.

In ^[27], a prediction technique for drone position is defined based on classification of drone data through the adoption of machine learning. Drone data captured at the ground controller are introduced to a naïve Bayes classifier to help predict the power utilization and current location of a drone, to potentially enable subsequent plans to continue or to interrupt drone flight. Data fields adopted for classification include drone altitude, switching status of the four transmitter coils and measured power transfer efficiency. Resulting drone position is compared against the actual drone position to verify the accuracy in classification. Training of the classifier is achieved through the introduction of past observations on drone flight trajectory, path and location as input to facilitate naïve Bayes model generation. Error rates in accuracy in the range 0.09% to 45%, were noted to depend upon the feature values such as the transmitter coil-switching values.

The authors in ^[28] proposed a methodology to detect the presence of a remotely operated drone, its current status and movement based solely on the communication between drone and the remote controller. They used random forest algorithms as the classifier. It also evaluates the effectiveness of the methodology in the presence of heavy packet loss and evasion attacks. The methodology is specifically designed and evaluated for remotely operated aircraft systems (RPAS) drones. They have shown a drone detection accuracy of 99.9% within 30 m without any packet loss and a detection accuracy >97% within 200 m with a packet loss up to 74.8%.

In ^[29], authors proposed UAV detection and identification based on radio frequency (RF) data using a hierarchical ensemble learning approach. The first classifier detects UAVs, the second one identifies the type of UAV, and the remaining two are used to identify the mode of operations. Each classifier used ensemble learning based on KNN and XGBoost algorithms. The proposed approach resulted in a classification accuracy of 99% with 10 classes. Each class uniquely identified the presence or absence of a UAV, its type (out of three different types of UAVs) and its mode of operation (ON mode, hovering mode, flying mode and recording mode). The paper also summarized the existing UAV detection using machine-learning approaches based on different data sources.

The authors in ^[30] provided a technique to identify the pilot of the drones based on radio control signals sent to a UAV using a typical transmitter. The dataset was collected from 20 different trained pilots flying the UAV through three different trajectories. The dataset consists of nine features including thurst, pitch, roll and yaw at time (t) and their derivatives at time (t). It also included control simultaneity variable at time (t) which describes the control signals available simultaneously at time (t). The proposed system used a random forest algorithm and resulted in an accuracy of 90%. The proposed technique can be used during forensic analysis to identify the pilot of the UAV and raise an alert in case of the suspected hijacking of a drone.

The authors in ^[31] proposed a methodology to detect drone status (flying or at rest) using just the encrypted communication traffic between the drone and the remote controller. The dataset was collected using communication from a drone running ArduCopter firmware. The encrypted packet information (without using its contents) was converted into six features (inter-arrival time, packet size, mean and standard deviation computed over a certain number of samples of inter-arrival time and packet size). Three different classifiers were used for classification (decision tree, random forest and neural networks). The random forest classifier provided better results for drone detection.

In ^[32], the authors identified the issue of inter-drone communication reliability, wherein transmitted packets may not reach the intended target successfully. The authors attempted to apply machine learning for accurate prediction of transmission patterns. The success/failure probabilities are computed using a Monte Carlo simulation setup comprising modeling channel design for transmission. The linear regression machine-learning technique was adopted alongside a comparative analysis with support vector machines (SVMs) with a quadratic kernel. The first property observed was the inverse proportionality between inter-drone distance and probability of a successful packet transmission. To foster measurement data collection, a total of 20 drones were simulated. Communication channel success in packet transmission was fixed at a 0.05 probability factor. Specific features identified for training of linear regression were transmission probability, node locations, transmission probability within a channel and time. For the SVM-QK classifier, features comprised quantization factor values, transmission probabilities, times, and locations of nodes in the network. Average prediction rates were found to yield a very low error rate of 0.00597

2.3. Machine Learning for Drone Forensics

There has been relatively less work on digital forensics for drones using machine-learning techniques. In ^[33], the authors conducted a survey on existing work in drone forensic domain (DRFs). They highlighted the challenges and opportunities in drone forensics. They also presented a methodology for drone-related event investigation. The existing work on forensic analysis for drone data has quite limited work focused on using machine-learning techniques for forensic analysis.

In ^[34], the authors proposed a methodology for drone forensic analysis and to identify suitable tools for performing digital forensics on drone data. Their work focused on the data acquired from the DJI Mavic air drone, and they compared three different tools including Airdata, CsvView and Autopsy. The acquired data included deleted files, attached devices information, emails and images along with audio and video files stored on the SD card. They presented their findings on the usage of these tools for forensic analysis and considered Airdata and Autopsy to be more suitable for drone data forensic analysis. The authors had earlier also provided a methodology based on a self-organizing map (SOM) for digital forensic analysis in ^[35]. The experiments were conducted using images acquired from ArduPilot DIY drone and DJI Phantom 4 drones. As part of the investigation, flight paths were extracted, and their associated datasets were obtained from both the drones. These datasets were further subjected to SOM-based clustering. The results obtained identify DJI Phantom 4 drones to hold more evidence and be forensically sound when compared with ArduPilot DIY drones.

In digital forensics, through clustering of common data samples into a single cluster and through subsequent visualization of the data clusters, commonalities between data elements can be observed, which can subsequently be labeled. Through the definition of such clusters for drones and the generated data during flight, it is possible to predict the trajectory of drones in flight and to label these as either legitimate flight paths or compromised ones that are typically exhibited by rogue/compromised drones ^[36]^[37].

Machine-learning techniques are beneficial in analyzing diverse datasets with variable volumes to generate inferences on the likelihood of an event occurring. Drone data analysis would benefit from such inferences as also evidenced in the recent literature that includes proposals to adopt machine-learning-based analysis of confiscated drones ^[24]^[25]^[28].

Table 1 provides a summary of the work presented in this section highlighting the different usages of machine-learning techniques for drone data and forensic analysis.

Table 1. Summary of machine-learning techniques used for drone data and forensics.

Paper	Year	Short Description	Scope	Evaluation Type
^[16]	2019	Machine-Learning Techniques used for UAV-based Communications	ML use to address communication concerns including channel modeling, resource management, positioning and security within UAV-based communication.	Survey Paper
^[21]	2017	Deep-Learning Techniques used in UAV problem domain	Survey paper highlighting work performed on use of deep learning for feature extraction, planning, situational awareness and motion control aspects of UAV systems based on Aerostack architecture.	Survey Paper
^[22]	2019	Rogue drone detection	A novel machine-learning approach to identify the rogue drones in mobile networks based on radio measurements	In Lab (Simulated data)
^[23]	2021	Deep-Learning-based technique for drone detection and identification using acoustic data	Uses CNN, RNN and CRNN-based architectures to identify drones using acoustic fingerprints of flying drones.	In Lab (Real data with augmentation)
^[24]	2018	Drone detection and identification system using Artificial Intelligence	Uses Haar classifier to detect a drone in an image and then uses CNN model to identify the type of drone.	In Lab (Real data)
^[26]	2020	Deep-learning-based anomaly detection for a vehicle in swarm drone system	The proposed anomaly detection model uses a deep neural network-based generation model to create a training model with normal data and perform tests with abnormal data.	In Lab (Real data)
^[27]	2017	Novel wireless power transfer system for drones using machine-learning techniques	Machine-learning model (using naïve Bayes) is used to identify position of the drone for enhancing wireless power transfer efficiency.	In Field
^[28]	2020	Drone detection via network traffic analysis	Detect presence, status and movement of drones by applying standard classification algorithms to the eavesdropped traffic, analyzing features such as packets inter-arrival time and size.	In Field
^[29]	2021	RF-Based UAV Detection and Identification	UAV detection and identification based on radio frequency (RF) data using hierarchical ensemble learning approach. The first classifier detects UAVs, second one identifies the type of UAV and the remaining two are used to identify the mode of operations.	In Lab (Real data)
^[30]	2018	Drone Pilot Identification based on Radio Control Signals	Describes an approach where radio control signal sent to an unmanned aerial vehicle (UAV) using a typical transmitter can be captured and analyzed to identify the controlling pilot using machine-learning techniques.	In Lab (Real data)
^[31]	2019	Detecting drones status via encrypted traffic analysis	Detect the current status of a powered-on drone (flying or at rest), leveraging just the communication traffic exchanged between the drone and its remote controller (RC) analyzing features such as packets inter-arrival time and size.	In Field
^[33]	2021	Research Challenges and Opportunities in Drone Forensics Models	It provides a detailed review of existing digital forensic models. It highlights the research challenges and opportunities through which an effective investigation can be carried out on drone-related incidents.	No evaluation
^[35]	2019	Digital forensics for drone data using SOM	Proposes a methodology based on self-organizing map (SOM) for digital forensic analysis of drone data.	In Lab (Real data)
^[32]	2016	Prediction of information propagation in a drone network by using machine learning	The packet transmission rates of a communication network with 20 drones were simulated, and results were used to train the linear regression and support vector machine with quadratic kernel (SVM-QK).	In Lab (Simulated data)
^[34]	2020	Digital Forensics for Drones: A Study of Tools and Techniques	Proposes a methodology that can help forensic investigators identify the most pertinent forensic investigation tools	In Lab (Real data)

This entry is adapted from the peer-reviewed paper 10.3390/su14084861

References

Renduchintala, A.; Jahan, F.; Khanna, R.; Javaid, A.Y. A comprehensive micro unmanned aerial vehicle (UAV/Drone) forensic framework. Digit. Investig. 2019, 30, 52–72.
Drone Technology Uses and Applications for Commercial, Industrial and Military Drones in 2021 and the Future. 2021. Available online: https://www.businessinsider.com/drone-technology-uses-applications (accessed on 14 April 2022).
Hummel, K.A.; Pollak, M.; Krahofer, J. A distributed architecture for human-drone teaming: Timing challenges and interaction opportunities. Sensors 2019, 19, 1379.
Yanmaz, E.; Quaritsch, M.; Yahyanejad, S.; Rinner, B.; Hellwagner, H.; Bettstetter, C. Communication and coordination for drone networks. In Ad hoc Networks; Springer: Berlin/Heidelberg, Germany, 2017; pp. 79–91.
Tanaka, H.; Matsumoto, Y. Autonomous Drone Guidance and Landing System Using AR/high-accuracy Hybrid Markers. In Proceedings of the 2019 IEEE 8th Global Conference on Consumer Electronics (GCCE), Osaka, Japan, 15–18 October 2019; pp. 598–599.
Yousef, M.; Iqbal, F. Drone forensics: A case study on a DJI Mavic Air. In Proceedings of the 2019 IEEE/ACS 16th International Conference on Computer Systems and Applications (AICCSA), Abu Dhabi, United Arab Emirates, 3–7 November 2019; pp. 1–3.
Bouafif, H.; Kamoun, F.; Iqbal, F.; Marrington, A. Drone forensics: Challenges and new insights. In Proceedings of the 2018 9th IFIP International Conference on New Technologies, Mobility and Security (NTMS), Paris, France, 26–28 February 2018; pp. 1–6.
Alhawiti, K.M. Advances in artificial intelligence using speech recognition. Int. J. Comput. Inf. Eng. 2015, 9, 1432–1435.
Nadkarni, P.M.; Ohno-Machado, L.; Chapman, W.W. Natural language processing: An introduction. J. Am. Med. Inform. Assoc. 2011, 18, 544–551.
Murphy, R.R. Introduction to AI Robotics; MIT Press: Cambridge, MA, USA, 2019.
Abdallah, A.; Maarof, M.A.; Zainal, A. Fraud detection system: A survey. J. Netw. Comput. Appl. 2016, 68, 90–113.
Kulik, S. Neural network model of artificial intelligence for handwriting recognition. J. Theor. Appl. Inf. Technol. 2015, 73, 202–211.
Voronin, V.; Marchuk, V.; Semenishchev, E.; Makov, S.; Creutzburg, R. Digital inpainting with applications to forensic image processing. Electron. Imaging 2016, 28, 1–7.
Francisca, O.; Ogbuju, E.; Alayesanmi, F.; Musa, A. The State of the Art in Machine Learning-Based Digital Forensics. SSRN Elec. J. 2020.
Bertino, E.; Kantarcioglu, M.; Akcora, C.G.; Samtani, S.; Mittal, S.; Gupta, M. AI for Security and Security for AI. In Proceedings of the Eleventh ACM Conference on Data and Application Security and Privacy, Virtual, 26–28 April 2021; pp. 333–334.
Bithas, P.S.; Michailidis, E.T.; Nomikos, N.; Vouyioukas, D.; Kanatas, A.G. A survey on machine-learning techniques for UAV-based communications. Sensors 2019, 19, 5170.
Hachimi, M.; Kaddoum, G.; Gagnon, G.; Illy, P. Multi-stage jamming attacks detection using deep learning combined with kernelized support vector machine in 5g cloud radio access networks. In Proceedings of the 2020 International Symposium on Networks, Computers and Communications (ISNCC), Montreal, QC, Canada, 20–22 October 2020; pp. 1–5.
Luo, P.; Wang, B.; Li, T.; Tian, J. ADS-B anomaly data detection model based on VAE-SVDD. Comput. Secur. 2021, 104, 102213.
Xiao, L.; Xie, C.; Min, M.; Zhuang, W. User-centric view of unmanned aerial vehicle transmission against smart attacks. IEEE Trans. Veh. Technol. 2017, 67, 3420–3430.
Hoang, T.M.; Duong, T.Q.; Tuan, H.D.; Lambotharan, S.; Hanzo, L. Physical layer security: Detection of active eavesdropping attacks by support vector machines. IEEE Access 2021, 9, 31595–31607.
Carrio, A.; Sampedro, C.; Rodriguez-Ramos, A.; Campoy, P. A review of deep learning methods and applications for unmanned aerial vehicles. J. Sens. 2017, 2017, 3296874.
Rydén, H.; Redhwan, S.B.; Lin, X. Rogue drone detection: A machine learning approach. In Proceedings of the 2019 IEEE Wireless Communications and Networking Conference (WCNC), Marrakesh, Morocco, 15–18 April 2019; pp. 1–6.
Al-Emadi, S.; Al-Ali, A.; Al-Ali, A. Audio-Based Drone Detection and Identification Using Deep Learning Techniques with Dataset Enhancement through Generative Adversarial Networks. Sensors 2021, 21, 4953.
Lee, D.; La, W.G.; Kim, H. Drone detection and identification system using artificial intelligence. In Proceedings of the 2018 International Conference on Information and Communication Technology Convergence (ICTC), Jeju, Korea, 17–19 October 2018; pp. 1131–1133.
CVOnline-Image Databases. Available online: https://homepages.inf.ed.ac.uk/rbf/CVonline/Imagedbase.htm (accessed on 14 April 2022).
Ahn, H. Deep learning based anomaly detection for a vehicle in swarm drone system. In Proceedings of the 2020 International Conference on Unmanned Aircraft Systems (ICUAS), Athens, Greece, 1–4 September 2020; pp. 557–561.
Jeong, S.; Bito, J.; Tentzeris, M.M. Design of a novel wireless power system using machine learning techniques for drone applications. In Proceedings of the 2017 IEEE Wireless Power Transfer Conference (WPTC), Taipei, Taiwan, 10–12 May 2017; pp. 1–4.
Sciancalepore, S.; Ibrahim, O.A.; Oligeri, G.; Di Pietro, R. PiNcH: An effective, efficient, and robust solution to drone detection via network traffic analysis. Comput. Netw. 2020, 168, 107044.
Nemer, I.; Sheltami, T.; Ahmad, I.; Yasar, A.U.H.; Abdeen, M.A.R. RF-Based UAV Detection and Identification Using Hierarchical Learning Approach. Sensors 2021, 21, 1947.
Shoufan, A.; Al-Angari, H.M.; Sheikh, M.F.A.; Damiani, E. Drone Pilot Identification by Classifying Radio-Control Signals. IEEE Trans. Inf. Forensics Secur. 2018, 13, 2439–2447.
Sciancalepore, S.; Ibrahim, O.A.; Oligeri, G.; Di Pietro, R. Detecting Drones Status via Encrypted Traffic Analysis. In Proceedings of the ACM Workshop on Wireless Security and Machine Learning (WiseML 2019), Miami, FL, USA, 15–17 May 2019; Association for Computing Machinery: New York, NY, USA, 2019; pp. 67–72.
Park, J.; Kim, Y.; Seok, J. Prediction of information propagation in a drone network by using machine learning. In Proceedings of the 2016 International Conference on Information and Communication Technology Convergence (ICTC), Jeju, Korea, 19–21 October 2016; pp. 147–149.
Al-Dhaqm, A.; Ikuesan, R.A.; Kebande, V.R.; Razak, S.; Ghabban, F.M. Research Challenges and Opportunities in Drone Forensics Models. Electronics 2021, 10, 1519.
Viswanathan, S.; Baig, Z. Digital Forensics for Drones: A Study of Tools and Techniques. In International Conference on Applications and Techniques in Information Security; Springer: Berlin/Heidelberg, Germany, 2020; pp. 29–41.
Mekala, S.H.; Baig, Z. Digital Forensics for Drone Data–Intelligent Clustering Using Self Organising Maps. In Future Network Systems and Security; Doss, R., Piramuthu, S., Zhou, W., Eds.; Springer International Publishing: Cham, Switzerland, 2019; pp. 172–189.
Fei, B.; Eloff, J.; Venter, H.; Olivier, M. Exploring forensic data with self-organizing maps. In Ifip International Conference on Digital Forensics; Springer: Berlin/Heidelberg, Germany, 2005; pp. 113–123.
Feyereisl, J.; Aickelin, U. Self-organizing maps in computer security. In Computer Security: Intrusion, Detection and Prevention; Hopkins, R.D., Ed.; University of Melbourne: Melbourne, VIC, Australia, 2009; pp. 1–30.

© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.