Ship-YOLOv5s: improved YOLOv5 ship target detection based on attention mechanism

The advancement of deep learning technology has greatly improved object detection in a variety of settings. The detection accuracy of multiple ship targets under a complicated background is still poor in the field of ships. In this paper, a SHIP-YOLOV5S model for multi-scale ship target detection and recognition is proposed by improving the YOLOv5 model. Especially the introduction of the Swin Transformer module, enabling the model to capture richer context and global information, and improve the detection performance of multi-scale ship targets. SimAM is simultaneously integrated into the model to enhance the detection accuracy of the model by highlighting the more important feature information and suppressing the less important feature information. So as to improve the detection accuracy of the model. The improved Ship-YOLOv5s model was compared and validated on the collection-produced ship dataset ShipData. The results show that the improved Ship-YOLOv5s model has a detection accuracy of 82.7%, a recall of 77.9%, and a mAP of 80.7% for ship targets, which are 2.1%, 1.0%, and 0.5%, respectively, compared to the YOLOv5 model. It shows that the improved model has more excellent ship target recognition and detection performance, which lays the foundation for subsequent applications in the fields of harbor management, navigation safety, and maritime search and rescue.