Intrusion Detection Method of Unmanned Aerial Vehicles Based on Lightweight Transformer

As unmanned aerial vehicle (UAV) technology develops rapidly and is widely used, illegal intrusion into UAV networks occurs frequently, posing a serious threat to public safety. To enhance the feature extraction ability of UAV network and effectively monitor the operation status of UAVs, we propose a lightweight Transformer network based on rotary position encoding, local aggregation attention unit, and lightweight feed-forward neural network for UAV traffic intrusion detection. Firstly, rotary position encoding is employed to encode the network traffic. To enhance the local and global information extraction of the network traffic, we propose a local aggregation attention unit, which can effectively aggregate and extract the local features by using group linear transformation, and realize the perception and enhancement of the global information by self-attention mechanism. Then, the enhanced features are extracted using a lightweight feed-forward neural network. Finally, the experimental results show that the proposed UAV traffic intrusion detection method has a more superior detection performance and provides a new lightweight solution for the security protection of UAV.