Research on Classification of Rail Defects Based on Image Processing Algorithm

There are many types and different shapes of rail defects. Even for the same type of defect, there are differences in the B-Scan images of the ultrasonic rail defect detection software. When the B-Scan image of a certain type of defect changes over a certain range, the detection software cannot identify this type of defect. Therefore, a classification for rail defects based on image processing algorithm was proposed. Firstly, the Tamura texture feature algorithm was combined with the local binary pattern algorithm to extract the feature values of the defect images, and form feature vectors. Secondly, the feature vectors of different kinds of defects were trained by support vector machine, and the optimal classification function was obtained. Finally, the category of untrained defects could be predicted by the optimal classification function. The experimental results showed that the proposed algorithm achieved high accuracy in the classification of rail defect images.