Detecting Flaws on Railways Using Semantic Segmentation

Abstract

Railway transportation usage is increasing day by day. However, as in all types of transportation, the safety of the road used in railway transportation is of great importance. In this study, an image classification-based approach is proposed to detect flaws on railway tracks. Two models have been proposed to detect flaws on railway tracks. In order to detect flaws on the railway, it is necessary to separate the pixels containing the flaws from the background images. Semantic segmentation methods are used in the literature to solve such pixel-based problems rather than class-based problems. The proposed models which are Unet and dilated convolutions have been successful in detecting flaws in railway tracks. The experiment of the proposed method achieved 99.99% success. © 2021 IEEE.