Effects of Faster Region-based Convolutional Neural Network on the Detection Efficiency of Rail Defects under Machine Vision

Abstract

To optimize the low efficiency of rail defect detection in China, a detection algorithm of Faster Region-based Convolutional Neural Network (Faster R-CNN) under machine vision is proposed in this research. First, the network structure of Faster R-CNN detection algorithm under machine vision, the realization of this algorithm, and the loss function and training steps are introduced. Then the structure of the proposed algorithm is studied. Finally, the network results of rail defect recognition and positioning and detection efficiency results of the algorithm are analyzed. The results show that the Faster R-CNN detection algorithm proposed in this research not only improves the detection accuracy, but also greatly optimizes the detection speed. In addition, the Faster R-CNN algorithm has a high accuracy rate in defect detection under machine vision. Accordingly, the results of this study are of great significance for realizing automatic surface defect detection of railway track in China.