Causal Inference-Enhanced UAV Detection and Identification for Low-Altitude Air City Transport

Recent progress in unmanned aerial vehicle (UAV) facilitates the promising low-altitude air city transport, which has high potential to alleviate traffic congestion but simultaneously triggers security-related issues. However, it is challenging to obtain reliable drone detection and identification (DDI) due to the uncertainties introduced by complicated urban environment, small inter/intra category differences, etc. In this work, we propose a causality-enhanced DDI method based on the radio frequency (RF) signal. First, causal chains are built with the signal’s statistical characteristics on time/frequency domain, quality index, and uncertainties (i.e., synchronization error and co-channel interference). Then, an agent-based modeling and simulation (ABMS) is utilized to obtain the parameters of those uncertainty factors. Finally, classification results of learning based method are further modified pairwise and online via the Bayesian network model (BNM), which is trained offline by the dataset with the constructed causal chains. In addition, we analyze the proposed method’s performance under different interferences, lightweighting strategies, etc. The proposed method has been deployed on edge device, whilst classification accuracy of vision-RF fusion has also been compared. Results of comprehensive experiments validate the superiority and effectiveness of the proposed causality-enhanced DDI method.