Segmentation of Railway Rails Using SAM

Abstract

Railways play a pivotal role in transportation infrastructure, but defects within them can lead to accidents and safety concerns. This article introduces an innovative approach for the automated segmentation of railway rails in images using the Segment Anything (SAM) model. SAM, a versatile image segmentation architecture, demonstrates the ability to segment objects without the need for extensive training data. In this study, the SAM model was applied to high-resolution railway images from the RailSem19 dataset, comprising 8,500 images. The results were remarkable, with an impressive, predicted Intersection over Union (IoU) score of 0.9760 for the left rail and 0.9595 for the right rail. This breakthrough offers a promising solution for enhancing railway safety and maintenance, showcasing SAM's potential to revolutionize image segmentation techniques in various applications. © 2023 IEEE.