An Improved Ship Detection Algorithm Based on YOLOv8 for SAR Images

Existing synthetic aperture radar (SAR) ship target detection algorithms are plagued by two primary issues: low detection accuracy and leakage. These issues stem from factors such as fuzzy target images, complex backgrounds, and a paucity of texture features of the target.To address these problems, this paper proposes a high-precision SAR ship target detection algorithm based on YOLOv8. The replacement of the C2f module in the YOLOv8 backbone with a CG-block module, which integrates local and global features, enhances the detection accuracy. This is due to the fact that the wide field of view of SAR images and the small size of ship targets necessitate an improvement in the detection accuracy.The neck part of the model is strengthened by the Gather-and-Distribute mechanism in the Gold-YOLO network, which improves the detection performance of small targets. The model incorporates the InnerSIoU loss function to enhance the regression accuracy, convergence speed, and model adaptability to complex scenes.Experimental results on the SSDD (Synthetic Aperture Radar Ship Detection Dataset) demonstrate that the enhanced algorithm attains a mean accuracy (mAP) of 98.42% and an accuracy of 96.21%, effectively achieving high-precision detection of ship targets in SAR images.