Corner-based ship detection method in complex SAR scenarios

Addressing the issue of low accuracy in ship detection using Synthetic Aperture Radar (SAR) images in coastal scenarios, this paper proposes a YOLOv10-based method. Firstly, SAR images are processed through the Harris algorithm to generate corner feature maps that reflect key scattering characteristics. Subsequently, the ECA(Efficient Channel Attention) is introduced to construct ECA-C2f module, thereby enhancing the focus of YOLOv10 on important channel-wise features. Finally, the SAR images along with the corner feature maps are input into the network to complete the detection. Furthermore, to verify the generalizability of corner features, this study also applies the feature across multiple detection networks. Experimental results show that the improved YOLOv10 achieves the AP50 of 94.11% on the test set, and the corner feature enhances accuracy across different detection networks. This study presents a high-precision method for ship detection in coastal SAR scenes and demonstrates the effectiveness and generalizability of corner feature in this task.