Mask R-CNN Architecture Based Railway Fastener Fault Detection Approach
Merve Yilmazer, Mehmet Karaköse
Abstract
Detecting and repairing faults in railway line components is of great importance in terms of transportation safety. Thanks to the successful results of deep learning techniques on images, progress has been made in defect detection studies. In this study, Mask R-CNN architecture, which enables segmentation in deep learning, was used to identify healthy and missing rail fasteners. Healthy and missing fasteners were labeled in the railway images obtained with the autonomous drone. The model was trained using labeled data and the performance of the model was evaluated with the data reserved for testing. It was determined that the method could detect healthy/missing fasteners with high accuracy rates and it was shown in the experimental results section. The precision value of the method is saved as 98% and the recall value as 96%.