An Investigation of Railway Fastener Detection Using Image Processing and Augmented Deep Learning

The rail fastening system forms an indispensable part of the tracks and needs to be periodically inspected ensure safe, reliable sustainable operations. Automated visual inspection has gained significant importance for fastener in recent years. Position accuracy, robustness, practical limitations due complex environment are some major concerns associated with this method. This study investigates combined use image processing deep learning algorithms detecting missing clamps within a system. images used was acquired during field inspections carried out along Borlänge-Avesta line Sweden. techniques proposed enabled improvement position removal redundant information from images. In addition, augmentation enhance data set, experimental reliability replicate challenges such inspection. Convolutional neural network ResNet-50 classification purposes, both achieved over 98% accuracy training validation 94% test stage. Both also maintained good balance between precision recall scores CNN were tested analyse their performances when clamp areas covered. able accurately predict state up 70% area occlusion, achieve accurate predictions 75% occlusion.