Semi-Supervised Framework for Rail Track Surface Damage Detection : Rail track surface damage detected by RTDS-Net

To achieve online high-speed and high-precision detection of track surface defects, this study collects the RSDD annotated public dataset and 2000 unannotated rail images as the model training and validation set. And an efficient high-speed railroad track damage detection network is proposed and trained by using a semi-supervised framework. The experimental results show that the detection accuracy of our method is 95.2% and the detection speed is 128 fps. The analysis results of the proposed model on the self-built dataset are compared with the analysis results of the expert in damage detection, which proves that the performance of the proposed model can satisfy the requirements of railroad damage detection.