In a modern urban railway transport system, monitoring the track conditions without interruption is the guarantee of secure and reliable operation of the system. Inspectors who follow traditional methods generally lack the ability to make timely inspections at all times, are inconsistent in their evaluations, and often fail entirely to discover the most subtly incipient defects. In this paper, an approach to deep learning-based railway track monitoring and fault detection using an image classification technique is proposed. To address these deficiencies, this paper proposes a framework for monitoring railway freight manner tracks and detecting faults that uses deep learning through image classification approaches. The proposed method utilizes Convolutional Neural Networks (CNNs) to automatically extract discriminatory features from railway track images prepared under a variety of frequently changing operating and environmental conditions. Four pretrained CNN architectures, NASNetMobile, MobileNetV2, DenseNet121, and EfficientNetB0, are implemented and compared to find the most suitable model for fault detection. Experimental results show that EfficientNetB0 outperforms the other models with an accuracy 98.11%, a precision 97.04%, a recall 99.24%, and an F1-score 98.13%. The proposed method’s reliability is further validated through thorough testing on unseen data. In addition, the optimized model is deployed using a Streamlit-based application so it can be interactively evaluated and visually demonstrated. This deployment illustrates the practical possibility of the suggested system for real-time railway track inspection, smart monitoring, and predictive maintenance applications.
Krishna et al. (Mon,) studied this question.