Ship Detection for SAR Remote Sensing Images based on Coordinate-Enhanced Attention and Long-Short-Range Spatial Semantic Fused Context

Due to the lack of sufficient appearance features of objects and the interference of complex background, it is a challenge to detect small-scale ships in SAR remote sensing images. To remedy these problems, we propose a small ship object detection algorithm by means of coordinate-enhanced attention mechanism and long-short-range spatial semantic fused context. Based on YOLOX, we introduce the spatial information into the channel attention to construct an attention-aware feature extraction network, which improves the focusing capability to small ships and suppresses background interference. The long-short-range spatial semantic fused context extraction module is used to capture the multi-scale environmental information around the objects with different receptive fields while integrating the overall semantic cues of the scene, so as to enhance the spatial and semantic feature representation of small ships. The experiments show that our method with the average precision 76.17% on LS-SSDD-V1.0 improves the detection performance of small-scale ships in SAR remote sensing images effectively.