Oriented Ship Detection Algorithm for SAR Images Based on Deep Learning

合成孔径雷达(SAR)图像船舶目标检测在民用和军事领域发挥着越来越重要的作用。然而，SAR图像中的船舶具有密集排列、任意方向和多尺度等特点。针对这些问题，文章提出了一种改进YOLOv11的SAR图像定向船舶检测方法YOLOv11-FM。首先，设计了一种快速混合聚合网络FMANet，增强网络的特征学习和提取能力。其次，提出了一种双向自适应特征融合网络BAFFN，通过跨层连接的方式实现更丰富的特征交互与融合。最后，采用小波特征增强模块WFU，改进颈部网络中上采样融合模块，增强船舶的细节信息。实验结果表明，YOLOv11-FM在RSDD-SAR船舶目标检测数据集上的P和AP50分别达到了94%和97.6%，具有良好的检测效果。Synthetic Aperture Radar (SAR) imagery for ship target detection plays an increasingly important role in civil and military applications. However, ships in SAR images are characterized by dense arrangement, arbitrary orientation, and multi-scale. To address these issues, this paper proposes a directional ship detection method, YOLOv11-FM, for SAR images with improved YOLOv11. Firstly, a fast hybrid aggregation network, FMANet, is designed to enhance the feature learning and extraction capability of the network. Secondly, a bidirectional adaptive feature fusion network BAFFN is proposed to realize richer feature interaction and fusion through cross-layer connection. Finally, the wavelet feature enhancement module WFU is used to improve the up-sampling fusion module in the neck network and enhance the ship’s detailed information. The experimental results show that YOLOv11-FM has a good P and AP50 of 94% and 97.6%, respectively, on the RSDD-SAR ship target detection dataset.