CAA-YOLO: an improved YOLOv11 model for ship detection in SAR images based on context anchor attention module

Protein Synthetic Aperture Radar (SAR) images overcome the limitations of traditional optical imaging technologies under adverse weather conditions and during nighttime. However, the diversity in ship scales poses challenges to the generalization capability of existing models, necessitating improvements in detection algorithms to enhance performance. In this work, we propose a detection method based on an improved YOLOv11. Specifically, the Neck structure of YOLOv11 is replaced with a Slim-neck module, which integrates the lightweight convolution GSConv to reduce computational complexity while improving accuracy. Furthermore, a Context Anchor Attention (CAA) mechanism is introduced to enhance long-range dependencies between pixels, enabling the model to better handle targets of varying scales. Experimental results show that the improved network achieves mAP50 and mAP50-95 values of 97.5% and 67.5%, representing increases of 1.2% and 2.0% compared to the YOLOv11s model. These enhancements significantly improve the effectiveness of ship target detection in SAR images.