面向SAR图像船舶目标检测的多尺度聚合扩散网络OA

Although object detection has made significant progress,most methods designed for natural images degrade sharply when applied to SAR ship images due to complex background clutter and multi-scale object variations.To address these issues,we propose DMADNet,a diffusion-based multi-scale detection network tailored for SAR ship detection.To enhance the detection accuracy,a Multi-scale Aggregation Network(MANet)is designed.Its core design concept is the independent feature extraction and interactive fusion strategy,which builds a flexible and efficient information flow processing framework,highlights the target features,and significantly improves the detection performance in the reasoning stage.Meanwhile,the Context Aggregation Attention(CA-X)designed is integrated into the network in a parallel manner,which can effectively integrate long-distance context information.While ensuring the help of the global context for target discrimination,it avoids the interference of irrelevant backgrounds,thereby significantly improving the detection accuracy of ship targets in complex backgrounds.DMADNet achieved mean Average Precision(mAP)scores of 96.65％,93.03％,and 97.92％on the SAR Ship Detection Dataset(SSDD),the High-Resolution SAR Image Dataset(HRSID),and the SAR-Ship dataset,respectively,under an IoU threshold of 0.5.These results further demonstrate the model's robustness and excellent detection capability in complex environments.