Fasteners Object Detection for Low-light Metro Undercarriage Environments

Aiming at the problems of low illumination, complex background, large scale differences among fasteners, and a majority of small-scale fasteners in the subway car underframe environment, this paper proposes a fastener target detection method for low-illumination subway car underframe environment. Firstly, the SCI image enhancement network is used to improve the image quality of the original images and restore various image information. Then, the improved RT-DETR target detection model is employed to detect fastener targets in the enhanced images. On the self-made subway car underframe fastener dataset, the [email protected]/% and [email protected]:0.95/% of the improved RT-DETR network reach 90.3% and 55.8%, respectively, which are 1.8% and 0.9% higher than those of the baseline network, and 5.2% and 3.0% higher than those of the dataset before image enhancement. Experiments show that the proposed method can effectively improve the detection accuracy of fasteners in low-illumination subway car underframe environment