SAR Ship Target Detection Based on Lightweight YOLOv5 in Complex Environment
Jiaqi Zhang, Jie Yang, Xuan Li, Zhenhong Fan, Zi He, Dazhi Z. Ding · 2022
In recent years, target detection methods based on depth learning have been widely used in the field of synthetic aperture radar (SAR) images, which is an indispensable part of the marine military field. Due to the particularity of SAR imaging system hardware platform, there are strict requirements for processor power consumption, which cannot meet the requirements of miniaturization. However, after the model is lightweight, the accuracy loss is often large. Therefore, to improve this common problem this paper compares the results of several YOLOv5 based on lightweight structure for ship target detection in complex scenes, and proposes an improved lightweight structure. The network performance is verified on our simulation dataset and SAR ship detection dataset(SSDD). The experimental results show that the lightweight structure proposed in this paper is superior to all lightweight structures at present in accuracy.