A two-stage algorithm of railway sleeper crack detection based on edge detection and CNN

The detection of sleeper crack is very important to ensure the reliability and safety of railway system. The powerful feature learning ability of convolutional neural network (CNN) can be used to detect cracks in sleeper images. However, convolutional neural network needs a lot of computation, which will reduce the detection speed. This paper presents a two-stage algorithm for the detection of railway sleeper cracks. In the first stage, the edge detection is carried out by using the 3 x 3 neighborhood range algorithm to find out the possible crack area, and a series of mathematical morphology operations are used to eliminate the noise target in the edge detection results. In the second stage, convolution neural network is used to classify the edge detected objects accurately. Through the analysis of many images of sleepers with or without cracks, it is proved that the cracks detected by 3 x 3 neighborhood range algorithm are coarser, clearer and more continuous than those detected by Sobel algorithm and Canny algorithm. In addition, the simple CNN model can achieve high image classification accuracy through edge detection and morphological operation of the railway sleeper crack image.