BCA-DetNet: Background Contrast-Based Attention Detection Network for UAVs Detection in Pulse-Doppler Radar

Radar serves as an effective tool for detecting unmanned aerial vehicles (UAVs). A variety of conventional approaches and deep learning-based methods have been developed to achieve effective target detection in radar systems. However, owing to the low, slow, and small characteristics of UAVs and the intricate background, those approaches face challenges in maintaining detection performance in low signal-to-clutter ratio (SCR) conditions. We found that the target occupies merely a minor portion of the range-doppler(R-D) map and accumulates a relatively small amount of energy, while the background clutter dominates the majority. Motivated by this insight, we propose a Background Contrast-based Attention Detection Network (BCA DetNet) for UAVs detection in pulse-Doppler radar, which assists object detection by learning background features. BCA-DetNet incorporates a random mask multi-scale background learner (RMBL) module to learn the distribution of background clutter. Subsequently, a background contrast-based attention (BCA) module is devised to suppress clutter in the feature dimension and the Jensen-Shannon (JS) divergence is introduced to the loss function to ensure the efficient operation of the BCA module. Ablation studies are conducted to explore the contributions of each modules. Experimental results on both synthetic and real datasets demonstrate the effectiveness of the proposed network.