A unified RF fingerprinting framework for device level UAV identification using HOCv2, wavelet RF-DNA, and deep learning

Tóm tắt

The rapid proliferation of unmanned aerial vehicles (UAVs) introduces significant challenges for spectrum surveillance and airspace security, particularly in device-level identification of identical platforms. This paper proposes a unified RF fingerprinting framework that integrates statistical features, multi-resolution time–frequency analysis, and deep learning representations. A multi-segment high order cumulant method (HOCv2) is introduced to preserve temporal dynamics, while wavelet based RF-DNA features capture multi-scale characteristics. These handcrafted features are fused to enhance discriminative capability. In parallel, time–frequency spectrograms are employed to train a lightweight convolutional neural network (CNN) for automatic feature extraction. Experimental results on seven identical UAVs show that feature fusion achieves 91.04% accuracy, while the CNN attains 92.37% accuracy with a macro AUC of 0.9949 and high computational efficiency. The results demonstrate the effectiveness and complementarity of the proposed unified framework for robust UAV RF fingerprinting.