SAR Ship Target Detection Based on Improved YOLOv5s
Yuankui Li, Xiaoqi Lv, Pingping Huang, Wei Xu, Weixian Tan, Yifan Dong · 2021 International Conference on Control, Automation and Information Sciences (ICCAIS) · 2021
Synthetic Aperture Radar (SAR) image ship target detection is of great significance to the field of marine monitoring. At present, many mainstream deep learning SAR image ship target detection networks have a large network model size, which is difficult to meet the requirements of equipment miniaturization. Therefore, in order to solve this problem, this paper proposes an improved YOLOv5s SAR image ship target detection network, which is based on the lightweight ideas of GhostNet and DWConv, and redesigns the structure of the YOLOv5s network model. The network performance is verified on the SAR Ship Detection Dataset (SSDD). Experimental results show that the model size of the improved YOLOv5s network proposed in this paper is only one-half of the original YOLOv5s, but its maen Average Precision(mAP) and Recall do not have much loss compared with the original YOLOv5s model.