A Ship Detection Method for SAR Images in Complex Scene Based on Improved YOLOv5

The traditional method of synthetic aperture radar image for ship detection is easily affected by complex background， resulting in the number increasing of missed targets. In this paper， the adaptive attention module composed of channel attention and spatial attention is employed for the YOLOv5 network. Most of network is allocated to target feature by using the weighted feature vectors in which the network’s ability of feature learning has been enhanced. Then the loss function of the detection model is optimized where the confidence of prediction boxes has been improved. Also the false negative probability has been reduced for target detecting in the complex background. According to simulations， the recall rate has been improved significantly by the proposed algorithm. The AP （Average Precision） value of ship detection for SAR image in complex scene has reached up to 79.8%， which increased by 26.1% and 17.3% respectively compared with the YOLOv5 algorithm and Faster R-CNN algorithm.