RS-YOLO: A Remote Sensing Ship Target Detection Algorithm Based on Feature Selection Alignment

With the continuous development of remote sensing technology, ship target detection technology based on visible remote sensing images has gradually become a research hotspot. To address challenges related to ship targets in remote sensing images with varying rotational angles and complex backgrounds, this study introduces the RS-YOLO (Rotate Ship-YOLO) algorithm, which is an enhancement of the YOLOv5 (You Only Look Once version 5) algorithm. The paper presents a multi-scale target detection approach utilizing the BiFaPN (Bi-directional Feature-aligned Pyramid Network) that integrates a feature selection module and a feature alignment module. Additionally, the study enhances the regression loss of YOLOv5 and incorporates the regression box WBF (Weighted Boxes Fusion) strategy to enhance the accuracy of the RS-YOLO ship target detector. Experimental results demonstrate that RS-YOLO achieves a 4.05% improvement in mAP0.5 compared to YOLOv5, showcasing its potential for application in remote sensing ship target detection.