SHIP-YOLO: A Lightweight Synthetic Aperture Radar Ship Detection Model Based on YOLOv8n Algorithm

Abstract

In response to the challenges posed by small objects, high noise, and complex backgrounds in synthetic aperture radar (SAR) ship detection, we proposed a lightweight model called SHIP-YOLO. In the neck of YOLOv8n, we replaced ordinary convolution (Conv) with a lighter ghost convolution (GhostConv) and introduced reparameterized ghost (RepGhost) bottleneck structure in C2f module. We then introduced Wise-IoU (WIoU) into the algorithm to improve the localization ability of the detection box. Finally, shuffle attention (SA) modules were added to the backbone and neck of YOLOv8n to enhance the perception capability of the target area. The results confirm that, compared with YOLOv8n, the proposed SHIP-YOLO on SAR Ship Detection dataset (SSDD) reduces the parameters and floating-point operations (FLOPs) by 17% and 11%, respectively, and improves the precision, recall, and mean average precision (mAP) (0.5) by 1.7%, 0.1%, and 0.2%, respectively. The proposed model also showed strong generalization ability on another Sar-Ship-Dataset.

Keywords

• Computer vision
• deep learning
• object detection
• radar remote sensing
• synthetic aperture radar