LERFSNet: A Lightweight SAR Ship Detection Model with Enhanced Receptive Field and Shared Decoupling Head

With the advancement of synthetic aperture radar (SAR) technology, SAR imagery has gained growing importance in fields such as natural disaster response, resource assessment, and military surveillance, particularly for ship detection. Current detection methods include traditional constant false alarm rate (CFAR) algorithms and the latest deep learning-based approaches. However, SAR ship detection continues to encounter challenges because of complex backgrounds, small target detection with significant scale variations, and limited computational resources. In response to these issues, this paper proposes a lightweight SAR ship detection model—LERFSNet. Built upon the lightweight general detection model YOLOX-tiny, LERFSNet incorporates a Large Effective Receptive Field (LERF) module and a Shared Decoupling Head (SDhead) to improve small target detection accuracy while reducing model complexity. Experimental outcomes on the SSDD and HRSID datasets demonstrate that LERFSNet offers notable improvements in detection accuracy and model parameter efficiency compared to existing methods.