WUSTCA: enhanced UAV RF signal classification with wavelet transform and STCA attention mechanisms

Unmanned aerial vehicles (UAVs) play a vital role in various civilian and commercial applications, necessitating accurate classification of their radio frequency (RF) signals. Current deep learning-based approaches struggle with high computational complexity, susceptibility to noise, and limited accuracy. This paper introduces a novel UAV signal classification framework that combines wavelet-based feature extraction with a hierarchical U-Net architecture, enhanced by split-time cross attention (STCA) and residual connectivity. The WUSTCA model effectively classifies UAV and controller signals, achieving an average classification accuracy of 96.6% for UAVs and 95.83% for UAV controllers on the CardRF dataset. By addressing challenges such as noise interference and signal diversity, this work provides a reliable and efficient solution for UAV signal classification, paving the way for real-time applications in complex environments.