ELSD-Net: A Novel Efficient and Lightweight Ship Detection Network for SAR Images
Shen Man, Weidong Yu · IEEE Geoscience and Remote Sensing Letters · 2025
Deep learning has shown significant promise in synthetic aperture radar (SAR) ship detection due to its powerful target recognition capabilities. However, SAR ship detection still faces unique challenges, including low image resolution, difficulty in detecting small targets, and the dense clustering of inshore ships. Additionally, traditional models tend to be complex and computationally intensive, making deployment on satellite platforms difficult. To address these issues, nonstride feature extraction (NSFE) module is proposed to construct the backbone network, significantly reducing the number of parameters, and multiscale efficient channel attention (MECA) is designed to further enhances the multiscale perception of the module. Moreover, strip partial convolution (SPC) module is designed to improve global modeling capability of the network with fewer parameters. Taking full advantage of the above modules and based on YOLOv5n architecture, we propose a novel efficient and lightweight ship detector named ELSD-Net for SAR images. Experimental results on the SAR ship detection dataset (SSDD) demonstrate that ELSD-Net achieves high detection accuracy 98.2% with fewer parameters 0.3 M and lower Flops 1.2 G compared to other methods.