Detection and Classification of Railway Track Surface Erosion-caused Holes and Scratches Defects Based on YOLOv5s

Abstract

Railway tracks are exposed to various environmental conditions, which can lead to defects that affect the safe operation of trains and potentially cause accidents. This paper presents an approach for detecting and classifying two common types of railway track surface defects, namely erosion-caused holes and scratches, using the YOLOv5s algorithm based on convolutional neural networks (CNN) and computer vision. The proposed method is compared with the YOLOv3-Tiny algorithm to demonstrate its superiority in terms of precision, recall, and mean average precision (mAP). A custom dataset containing labeled images of both defect types was collected and used for training and evaluation. The experimental results show that YOLOv5s outperforms YOLOv3-Tiny. The improved performance of YOLOv5s can be attributed to its optimized architecture and better balance between speed and accuracy. The limitations of the study include the relatively small dataset size and the need for further validation in real-world scenarios. Future work will focus on expanding the dataset, investigating advanced data augmentation techniques, and integrating the model with other sensing modalities for robust and real-time defect detection in railway maintenance systems. The proposed approach has the potential to significantly improve the efficiency and effectiveness of railway track inspection and maintenance processes.