•  
  •  
 

Abstract

Maritime ship detection is of great significance for both military security and civilian applications. Synthetic Aperture Radar (SAR), with its all-weather and all-day imaging capability, plays a vital role in maritime surveillance. Nevertheless, SAR ship targets typically appear small in scale, embedded in complex backgrounds, blurred at boundaries, and easily confused with near-shore features, which pose substantial challenges for accurate detection. To address these issues, we propose a SAR ship detection network that integrates dual enhancements of small-object representation and edge information. The network introduces two key components: the Small Target Refine Pyramid (STRP) to strengthen shallow feature representation for small targets, and the Edge Information Enhancement and Propagation Mechanism (EIEP) to refine and transmit boundary cues. Additionally, a customized bounding box regression loss, Focaler-MPDIoU, is designed to improve localization accuracy and robustness to complex object morphologies. Experiments on benchmark SAR ship detection datasets (HRSID, SSDD, and LS-SSDD) demonstrate that the proposed method achieves superior performance compared with both classical and state-of-the-art detection models.

Abstract Image

Download
COinS