Road Trash Detection Algorithm of RT-DETR Integrated with State Space

As global urbanization progresses rapidly, the issue of urban road trash is becoming increasingly severe. Traditional manual cleaning methods are inefficient and struggle to meet the growing demand. This paper proposes an improved road trash detection algorithm based on RT-DETR (Real-Time Detection Transformer) and State Space Model (SSM) to enhance detection accuracy and robustness. The algorithm incorporates an innovative feature pyramid structure and a state-space-based internal scale feature enhancement module, effectively addressing object detection in complex scenes as well as challenges associated with occlusions and small object detection. We constructed a diverse urban road trash detection dataset comprising over 60,000 images, providing a solid foundation for the training and testing of the algorithm. Experimental results show that the proposed model improves the mean average precision (mAP) on the COCO dataset by 0.8 percentage points and achieves a 1.2 percentage point improvement on our custom road trash detection dataset, with a 2.1 percentage point increase in the precision of small object detection. The proposed algorithm achieves a favorable balance between real-time performance and accuracy, significantly enhancing trash detection in complex environments, particularly in terms of small object recognition and detection accuracy under occlusion conditions.