Hierarchical Deep Learning Framework for Enhanced UAV Classification Mitigating Bluetooth and WiFi Interference

In recent years, there has been a widespread use of unmanned aerial vehicles (UAVs) or drones on both commercial and defense applications. There is an increasing security concern as these UAVs can also be used for illegal activities by malicious users. It is necessary to detect and classify these malicious UAVs or drones in order to neutralize them. In this paper, a hierarchical scheme is proposed for detecting and classifying UAVs based on the RF signatures obtained from the RF link (both uplink and downlink) that is used to control the UAV from the ground station or UAV controller. The work also addresses the challenge of classifying UAV RF signatures in the presence of interferences like WiFi and Bluetooth as they also operate in the same frequency band. The method involves entropy-based steady-state extraction, segmentation, and spectrogram image conversion followed by a hierarchical classification based on different variants of convolutional neural network (CNN). The proposed method is validated on signals taken from the CardRF database and achieves an average detection accuracy of 99.7% and an average precision of 99.93%. As it is a hierarchical classification scheme, performance measures at each hierarchical stages are reported in the simulation results.