SFRT-DETR:A SAR ship detection algorithm based on feature selection and multi-scale feature focus

Currently, ship images acquired by synthetic aperture radar (SAR) are susceptible to complex marine environments and inconsistent ship sizes, which brings great challenges to lightweight, high accuracy, and real-time SAR ship detection. To address these issues, we propose the SFRT-DETR algorithm: a SAR ship detection algorithm based on feature selection and multi-scale feature focus. Firstly, the feature selection module is designed to screen the SAR ship image features through the attention mechanism, so this module can filter the redundant background feature information and improve the detection speed of the model. Then, a multi-scale Feature Focus (MFF) module is constructed, which uses parallel dilated convolution to capture and focus ship features at different scales. This module effectively improves the ability of the model to detect ships of large, medium, and small sizes. Finally, multi-path up-sampling and down-sampling modules are constructed, which can enhance more meaningful multi-scale ship features. The experimental results on the High Resolution SAR Images Dataset (HRSID) and SAR Ship Detection Dataset (SSDD) demonstrate that, in comparison with the baseline RT-DETR model, the Average Precision (AP) has been enhanced by 5.37% and 7.30%, respectively. The Frames Per Second (FPS) has been elevated to 77 frames/s, thereby achieving the goal of balancing high accuracy with real-time detection.