Research on Ship Detection Algorithm in SAR Images Based on YOLOv8
Zibo Liu, Wenke Hou, Jinghang Wang, Zhiguo Zhou · 2025
SAR (Synthetic Aperture Radar) is a remote sensing technology based on the principle of radar active imaging, which generates high-resolution images of detected targets by sending and receiving microwave signals. Compared with other remote sensing technologies, it has the characteristics of all-day, all-weather, and no dependence on visible light and weather conditions. However, due to the unique imaging characteristics of SAR images and the fact that ships are often densely distributed as small targets, algorithms that perform well in optical image detection frequently encounter numerous false positives and missed detections in SAR ship detection. Therefore, this paper proposes a lightweight SAR image ship detection algorithm based on higher-order spatial interaction mechanism, which introduces SCConv convolution to achieve lightweight optimisation, and improves higher-order spatial interaction and salient feature capture capability through recursive gated convolution and NAM attention mechanism, in addition to the introduction of Box-Cox variations to control the loss function through the power parameter$\alpha$to control the loss function for different IoU intervals, to improve the Bounding box localisation accuracy. The experimental results show that the algorithm achieves excellent results in detection accuracy, detection speed and number of parameters, which is of great application value for remote sensing ship detection research.