A Semi-Supervised Unmanned Aerial Vehicle Recognition Method Based on Self-Distillation

The widespread use of the unmanned aerial vehicle (UAV) poses a threat to public safety. Therefore the recognition of UAV becomes more and more important. Existing deep learning based UAV signal recognition methods rely on a large number of labeled samples and perform poorly when the labeled training samples are small. To solve these problems, this paper proposes a framework for UAV recognition based on self-distillation. And a vision transformer (ViT) based time-frequency encoder is proposed to extract the features of UAV signals. The proposed method can make good use of unlabeled samples and thus maintains better performance when there are fewer labeled samples. Simulation results show that the recognition accuracy of our proposed method with fewer labeled samples is better than the recently reported works.