A Lightweight UAV Recognition Method Based on Multi-Feature Fusion of Micro-Doppler Images

To address the growing complexity of airspace management resulting from the widespread use of unmanned aerial vehicles (UAVs), this paper presents a lightweight UAV recognition method based on multi-feature micro-Doppler image fusion. Considering millimeter-wave radars, the proposed method enhances the micro-Doppler features by applying a singular value decomposition (SVD)-based approach to eliminate the target’s main body velocity from the time-frequency spectrum. The enhanced features including the cadence-velocity diagram, short-time cepstrum, are then integrated into multi-dimensional grayscale fusion. Meanwhile, an improved MobileNetV3 is proposed for efficient recognition and low complexity implementation by introduction an efficient channel attention (ECA) module. The proposed method delivers high classification accuracy while maintaining a lightweight network architecture, making it well-suited for resource-limited embedded systems. Experimental results demonstrate the computational efficiency and effectiveness of the approach for UAV recognition compared to the state-of-the-art methods.