Nearshore Ship Detection on SAR Image Based on Yolov5
Qiang Fu, Jie Chen, Wei Yang, Shichao Zheng
Abstract
Nearshore ship detection faces big challenge due to the missing alarms and false alarms caused by onshore ship-like objects and close arrangement of ships. This paper proposes a method to detect nearshore ships, which is based on You Only Look Once Version 5 (Yolov5). To improve the precision, attention model and Circle Smooth Label (CSL) are unified into the detection network. The main research content and experimental work of this paper are as follows. First, Yolov5 network, attention model and CSL algorithm are analyzed. After that, the detection experiment is carried out based on Yolov5. Next, the attention model is introduced to improve the network. Then, combined with the CSL algorithm, the Yolov5 rotation detection network is reconstructed. Finally, by adjusting the training parameters and improving the attention, the test result of the detection network for inshore targets reached mAP above 80%, and the feasibility of the CSL+Yolov5 algorithm to achieve rotation detection is confirmed.