Detection of Rail Surface Defects with Two Deep Learning Methods: Comparative Analysis

From past to present, railway transportation is frequently preferred by people due to its economic and safe nature. TCDD personnel carry out the soundness control of the railway tracks by visual inspection. Visual inspection takes a long time. Performing the inspection process of the rails autonomously by the computer will speed up the detection of defective rails and minimize the risk of railroad accidents. In this study, automatic detection of defects in the rails is provided with the CNN architecture. In the study, a total of 2000 images belonging to two classes, broken and intact, were used. With the GoogleNet CNN architecture, the ray images were classified correctly at a rate of 96.5%. The study can help businesses in the inspection of the rails for the safety of railway transportation.