In SAR image ship-target detection scenarios, due to small-target characteristics and susceptibility to background interference, the detection effect of ship targets is poor. To reduce false and missed detections of ship targets and to improve target detection accuracy, we come up with an improved YOLOv5 detection model. On the basis of YOLOv5 algorithm, we use Swin Transformer as a feature extraction module in Backbone to improve multi-scale object detection capability while reducing computational complexity. Based on FPN in the Neck section, we have constructed a new feature fusion module which allows for more comprehensive feature fusion. In head section, a detection head is added to improve the detection capacity of small targets. In this paper, comparing with the YOLOv5s model on the SSDD dataset, the parameters P, R, mAP@.5, and mAP@.5:.95 of the proposed model have been improved by 1.133%, 0.416%, 0.181%, and 2.871% respectively.