SAR ship detection network based on global context and multi-scale feature enhancement

With the rapid development of synthetic aperture radar (SAR) technology, SAR image ship detection plays a crucial role in fields such as marine environment monitoring and maritime traffic control. Due to the large-scale difference between ship targets and the small spatial feature information of targets in complex environments, the existing network models have low detection accuracy and cannot achieve satisfactory performance. Therefore, in this paper, a SAR ship detection network (FGNet) based on global context and multi-scale feature enhancement is proposed. The method adopts the global context block (GC block) to effectively model the global context information of the feature map and enhance the network’s focus on important features. In addition, we propose the feature enhancement module (FEM), which enhances the network’s feature representation capability for ships of different scales by designing a multi-branch structure with varying receptive field sizes. Experimental results on the SAR-Ship-Dataset and SSDD datasets show that the average accuracy of our method is 95% and 96.58%, respectively. Compared with other ship detection methods, the method has high detection accuracy and excellent generalization performance.