SD-YOLO: High-precision network for SAR ship target detection

Recently, synthetic aperture radar (SAR) images ship target detection for marine security plays a vital role. While, the existing target detection methods have some problems, such as low precision, poor real-time, lack of detection and false detection. To improve the precision without reducing the speed, a target detection method SD-YOLO for SAR ship detection is proposed. SD-YOLO consists of a backbone, a neck, and a head. The backbone module is mainly used to extract feature information, while the neck module is mainly used for feature fusion to facilitate the follow-up detection of the head module. The neck module mainly uses the fused neck information to carry out multi-scale detection and generate the result map. We tested the effectiveness of our approach on the SSDD dataset. Our experiments show that the model performs better in the case precion, efficiency and model complexity compared to the state-of-the-art (SOTA) model You Only Look Once version 8 (YOLOv8) model.