In order to improve the efficiency of train wheel tread detection, a dynamic wheel tread detection system based on machine vision was designed, and k-means clustering algorithm was analyzed. The algorithm was improved by weighted Euclidean distance. With the clustering method’s characteristics of maintaining maximum similarity, k-means clustering algorithm based on weighted Euclidean distance was used for image processing in machine vision. First, the original image was preprocessed by image enhancement and gray scale, and then the image was threshold segmented using feature clustering to make the features of each part of the image more prominent. The image processing results show that the visual detection system of wheel tread damage based on weighted Euclidean distance k-means clustering algorithm can effectively detect tread damage.