A Ship Target Intelligent Detection Algorithm Based on Improved YOLOv5s

Currently, the deep learning-based synthetic aperture radar (SAR) image target detection methods have limited detection ability on small targets, and their accuracy needs to be improved either. In order to improve the method’s recognition ability and accuracy on small targets, this paper introduced the coordinate attention mechanism to obtain the long-range dependency relationship based on YOLOv5s model, and used SIoU loss function to improve the detection speed and accuracy of the model. The optimized YOLOv5 algorithm are significantly improved the performance and calculation speed of the detector. The experimental results on the High-Resolution SAR Image Dataset (HRSID), indicated that the accuracy and recall rate of the proposed method are increased by 2.566 \(\%\) and 0.871 \(\%\) , respectively. At the same time, mAP0.5 is increased by 0.579%, mAP0.5-0.95 is increased by 2.737 \(\%\) , the training time is significantly reduced, and the detection performance is greatly improved.