Waysides Inspection Using Wayside Processing Imaging and Deep Learning
R. Thillaikkarasi, M. Mohamed Yaseen, R. Rameshbabu, Radha Prabhakaran, R. Kesavan, Jose Anand A.
Abstract
Advances in deep learning techniques have made computer vision tasks more accurate and faster by relying on convolutional neural networks and more powerful hardware. In industry, automatic inspection supported by these methods is capable of offering constant maintenance and avoiding accidents on railways. Thus, this work proposes the application of deep learning and image processing methods to perform the automatic inspection of wheel sets in train cars. More specifically, the size of the wheel and the thickness of the bandage are measured, in addition to locating the bearings' fixation screws. The constructed neural network performs semantic segmentation on photographs provided by the mining company Vale. Using a U-Net architecture, with ResNet50 as a backbone, the network was able to reach 92.50% in mIoU and 97.52% in mPA, metrics adopted to evaluate this proposal. The post-processing step recovered the screws and improved the evaluation metrics, indicating the success of the proposed inspection.