GS-DETR: Accurate and Efficient Object Detection in UAV Imagery with Gated Feature Fusion and an Enhanced Pyramid

Object detection in unmanned aerial vehicle (UAV) remote sensing imagery remains a significant challenge due to complex backgrounds, multi-scale objects, and a high prevalence of small targets. To address these issues, a Gated Feature Fusion Net and Small Object Detection Pyramid (GS-DETR) is presented, based on the Real-Time Detection Transformer (RT-DETR) framework. Specifically, a gated feature fusion net is designed to reduce model parameters. It features a lightweight backbone that enhances target feature representation through a unique gating mechanism and anisotropic feature extraction. Furthermore, a small object detection pyramid (SODP) is implemented to preserve high-resolution details and integrate a mini-kernel that utilizes a lossless down-sampling module for deep feature optimization. This design allows the framework to achieve a superior balance between detection accuracy and model efficiency while maintaining the real-time capabilities of its baseline. Extensive experiments on the VisDrone2019 and CoDrone datasets demonstrate that, compared to the baseline model, GS-DETR improves mAP@0.500 by 2.3% and 1.3%, respectively, while reducing the parameter count by 11.3%.