SCueU-Net: Efficient Damage Detection Method for Railway Rail
Jun Lu, Bo Liang, Qujiang Lei, Xiuhao Li, Junhao Liu, Liu Ji, Jie Xu, Weijun Wang
Abstract
Automatic detection of industrial product damage using machine learning is a promising research area. At the same time, various machine learning methods based on convolutional neural networks have a very important role in the application of visual automatic detection. Therefore, the machine vision-based automatic detection of high-speed railway rail damage has received widespread attention. This paper proposes an efficient detection method for the damage of high-speed railway rails called SCueU-Net. For the first time, the combination of U-Net graph segmentation network and the saliency cues method of damage location is applied to the task of high-speed railway rail damage detection. The experimental results show that our method has a detection accuracy rate of 99.76%, which is 6.74% higher than the recent method in damage identification accuracy, which improves the detection efficiency of rail damage significantly.