Synthetic Aperture Radar (SAR) ship detection is crucial for maritime traffic management, search and rescue, and environmental monitoring but remains challenging due to small targets, blurred contours, and complex ocean backgrounds. To address these issues, this paper proposes YOLO-BDM, an improved detector based on YOLOv11. The Diverse Branch Block (DBB) is introduced into the backbone to enhance feature representation through multi-branch training and reparameterized inference. A Multi-scale Contextual Attention (MCA) mechanism is integrated into the backbone and neck to strengthen multi-scale semantic modeling and background discrimination. Additionally, a four-layer Bidirectional Feature Pyramid Network (BiFPN) is employed for efficient multi-scale feature fusion. Experiments on the SAR-Ship dataset show YOLO-BDM achieves 97.27% mAP, 94.11% Precision, and 93.07% Recall, surpassing the baseline and validating its effectiveness.
Xiong et al. (Thu,) studied this question.