EES-DETR: A Transformer-Based Detector for Challenging Small Object Scenarios

Small-object detection for unmanned aerial vehicles (UAVs) remains challenging in complex environments due to extremely small targets (often smaller than 10 pixels), frequent occlusions, and background interference. Existing methods struggle to capture fine-grained details of small objects while maintaining a balance between detection accuracy and computational efficiency. To address these challenges, we propose EES-DETR, an end-to-end Transformer-based framework tailored for small-object detection. The RT-Block backbone enhances sensitivity to object contours and fine details by synergistically extracting edge-spatial information through SobelConv-based edge-spatial feature fusion. In addition, the FC-Fusion module introduces a focus-diffusion mechanism that improves multi-scale small-object detection via cross-scale feature interaction and contextual information propagation. Experimental results on the VisDrone2019 dataset show that EES-DETR improves AP and AP50 by 3.1% and 4.3%, respectively. These improvements significantly enhance detection performance in complex aerial scenes, demonstrating superior precision and localization accuracy for real-time UAV applications.