Small Ship Detection Method of SAR Images Using Lightweight Feature-fused Enhancement Learning

The low resolution and complex environment of small targets in synthetic aperture radar (SAR) ship images make it difficult to extract small target. Therefore, it is prone to the false detection and missed detection. Addressing these issues, this paper introduces a SAR ship small target detection algorithm based on YOLOv7. Firstly, To tackle the problem of insufficient feature information and vulnerability in small target detection, a novel approach named Multi-Scale Convolution with Convolutional Block Attention Module (MSCBAM) is proposed. This approach aims to enhance feature representation, enabling the model to capture more comprehensive feature information for small ships. Secondly, since the shallow feature map contains more detailed information, the shallow detection branch is added to reduce the loss of small target information. Finally, in order to realize the model more lightweight, the Group Separable Convolution (GSConv) is introduced in the neck model. Experiments on the SSDD data set show that the proposed algorithm can effectively improve the detection ability of small targets. The average accuracy mAP reaches 95.6 %, which is 2.2 % higher than the original YOLOv7, and the small target detection accuracy APs reaches 65.5, which is 5.5 % higher than the original YOLOv7.