FDSS-YOLO: A More Accurate Remote Sensing Object Detection Method

To address the challenges of complex backgrounds, small targets, and low detection efficiency in remote sensing images, this paper proposes an improved FDSS-YOLO method. The method introduces a Coordinate Attention (CA) module to focus on key regions, and outputs a 160×160 pixel shallow feature map to retain more fine-grained features of small targets, thereby improving small object detection performance. Additionally, a spatial-channel decoupling down-sampling layer (SCDown) is integrated to reduce redundant parameters while enhancing detection accuracy. Through lightweight design, the improved model achieves a better balance between speed and accuracy. Experimental results show that compared to the baseline model, the improved model achieves a 4.8% and 1.2% increase in mAP@50 on the RSOD and NWPU VHR-10 datasets, respectively, and a 7.5% and 1.4% increase in mAP@50:95, validating the effectiveness and applicability of FDSS-YOLO.