Classification of Rail Track Crack using CNN with Pre-Trained VGG16 Model

Abstract

One of the vital components of railway infrastructure is rail tracks. Maintenance of rail track has been a major challenge in most of the countries and one such challenge is the detection of cracks on the rail surface. To maintain good health of the tracks requires regular inspection and prompt action, failure to which, may lead to accidents and loss of lives. The railway department is introducing many innovative methods to make the inspection process efficient. In the past, various methods have been explored to detect defects on rail surfaces such as Computer Vision-Based method, but full automation is far from achievement. Few of the advanced countries are making use of Deep Learning techniques to monitor and maintain the condition of rail tracks. In, this paper, amalgamation of Convolutional Neural Network (CNN) and transfer learning is applied for classifying defective (with cracks) and non-defective rail surfaces.