Towards SAR Ship Detection with Semantic Mining and Small Scale Detector

With the development of China's shipping industry, effective maritime management requires a sound monitoring system. This article improves YOLOv11 for ship detection in lightweight SAR images, with a focus on small ship targets and complex port environments. We added a visual converter to the feature extraction network to enhance semantic understanding, introduced a small object detection layer with additional detection heads to better capture small object features, and designed a custom loss function for sparse backgrounds based on SIoU. Our experimental results on SSDD and HRSID datasets demonstrate that, our method improved accuracy by 2.5% and recall by 6.1% compared to YOLOv11, with only an increase of 0.2MB. This method significantly improves the ship detection and maritime monitoring capabilities.