An Improved Model Based on RT-DETR for Remote Sensing Object Detection

Remote sensing image object detection is a critical task in computer vision, particularly in high-resolution images where large-scale variations and complex backgrounds present significant challenges. Existing methods encounter bottlenecks in accuracy and efficiency when applied to these tasks. To address these challenges and enhance detection performance, this paper proposes an improved algorithm based on the real-time detection transformer (RT-DETR). We introduce the FasterRepBlock, Slim-neck, and the Powerful-IoU bounding box regression loss function. By enhancing feature extraction, multi-scale fusion, and bounding box localization, we improved both detection accuracy and computational efficiency. Experiments on the high-resolution remote sensing object detection dataset (TGRS-HRRSD) demonstrate that the proposed method significantly improves both detection accuracy and computational efficiency. Compared to mainstream models, the proposed method achieves superior performance on common object detection metrics and effectively handles remote sensing image detection tasks.