Detection and Classification System for Rail Surface Defects Based on Deep Learning

Abstract

AbstractThe detection and classification of rail track surface defects is critical to the safety and productivity of urban transport operations. The requirement of fast and accurate detection and classification from the huge number of images taken drives for an automated solution. A method for detection and classification of rail surface anomalies and defects adopting deep learning approach based on the convolutional neural networks has been presented in this paper. The training and testing images are acquired from an automated video recording setup on the train. We proposed a convolutional neural network model trained to learn the features which will then be utilized to detect and classify defects from images taken by image acquisition device mounted on the train. The experimental results are promising and can be integrated to the current workflow of rail maintenance operations to improve the productivity.KeywordsDeep learningConvolutional neural networkRail surface defectDefect detectionDefect classification