Automated Railway Crack Detection Using Machine Learning: Analysis of Deep Learning Approaches

Detecting defects in railway tracks, particularly small cracks or gaps, is traditionally a labor-intensive task. By leveraging machine learning, this process can be automated and accelerated, reducing both time and costs. This paper evaluates several prominent deep learning models for identifying railway cracks and examines the key factors influencing the training process. While our primary focus is on various versions of the You Only Look Once (YOLO) object detection model, we also explore the Residual Networks model. Our findings indicate that YOLOv5 and YOLOv9 achieve high crack detection accuracy, with $\mathbf{F 1 - s c o r e s}$ of $\mathbf{0. 9 2}$ and $\mathbf{0. 9 1}$, respectively. These results underscore the efficacy of deep learning models in detecting and classifying cracks, thereby potentially lowering labor costs. Additionally, we employ Explainable AI techniques to elucidate the models’ decision-making processes in crack detection. By automating the detection of railway track defects, we can significantly enhance the efficiency and reliability of railway maintenance. The use of deep learning models, particularly the YOLO series, has shown promising results in accurately identifying even the smallest cracks. This not only speeds up the inspection process but also ensures a higher level of safety by detecting potential issues before they become critical. The integration of Explainable AI techniques further adds value by providing insights into how these models make decisions, which is crucial for gaining trust and improving the models. Overall, our research highlights the potential of advanced machine learning techniques to revolutionize railway maintenance, making it more cost-effective and reliable.