Multiclass Classification of Rail Track Defects Using Deep Learning Techniques

In India and around the world, the most preferred transportation is by rail because of the comfort, it provides for passengers and its reasonable pricing. India has one of the largest rail networks, and its major responsibility lies in maintaining these rail tracks in an upright condition. The maintenance task comprises the identification of different types of defects on the tracks and taking corrective action for the defects found. This research work is aimed at automating the process of identifying defects on the tracks using Deep Learning techniques. Seven types of defects are considered. Convolutional Neural Network (CNN), Xception, and MobileNet models are trained on the dataset consisting of 790 images. Results show that the MobileNet model gives maximum accuracy of 91 % for the test data.