Improvement of Multi-Scale Ship Detection Algorithm for SAR Images Based on YOLOv7

In recent years, deep learning, especially Convolutional Neural Networks (CNNs), has made significant progress in ship target detection in Synthetic Aperture Radar (SAR) images. However, existing methods still face challenges such as missed detections and false positives when dealing with complex maritime scenes and multi-scale ship targets. To address these issues, this paper proposes an improved YOLOv7 model—ACF-YOLOv7, focusing on multi-scale ship detection in complex environments. This study designs an innovative CB-BiFPN module that combines the Bidirectional Feature Pyramid Network (BiFPN) and Convolutional Block Attention Module (CBAM). By introducing a learnable weighted mechanism, this module efficiently fuses features, enhances the focus on key features, and suppresses redundant information. It effectively reduces feature loss for small ships and fully integrates ship target feature maps across different scales. Additionally, the paper introduces the SIoU loss function to further optimize target localization accuracy. Experiments on the SSDD dataset show that ACF-YOLOv7 achieves improvements of 2.6%, 1.8%, and 3.5% in detection accuracy (94.11%), recall rate (92.24%), and mean average precision (mAP), respectively, compared to the baseline YOLOv7. Comparative experiments also demonstrate that ACF-YOLOv7 outperforms other deep learning methods in complex scenarios. This research provides an efficient and robust solution for ship target detection in SAR images in complex environments.