UAV Attack Detection and Mitigation Using a Localization Verification-Based Autoencoder

Interest in the security aspect of unmanned aerial vehicles (UAVs) has intensified in recent years due to the increased adoption of UAVs in various civilian and military applications, even though their built-in security may be ineffective against control and spoofing attacks. In this paper, we propose a technique that exploits the unique nature of the physical layer attributes in such systems to enhance their security capabilities. Specifically, we use the received signal strength (RSSI) and angle of arrival (AoA) to build a localization-based verification system to authenticate the received signal. To this end, we adopt a deep learning autoencoder to classify events based on trained and optimized parameters. The generated dataset contains the geographical coordinates and attributes of the received signal for each sample in the UAV’s maneuvering path. To verify the legitimacy of the received signal, the algorithm specifies a reconstruction loss threshold. The results show that high accuracy, as represented by a high F1 score, is achievable with this approach. We compare the proposed detection system with a benchmark with or without a smart attack in terms of false alarm rate. The proposed system outperforms the benchmark in both cases.