Application of VGG-16 and modified VGG-16 for railway fault classification using image datasets

Railway accidents, despite safety measures, remain a significant concern, stemming from collisions or derailments due to technical defects (rolling stock failure, rails degradation, safety systems malfunction) and external factors like terrain movements or obstructions. Such events have severe consequences, including injuries, fatalities, transportation disruptions, and environmental damage. This study focuses on enhancing railway safety by detecting and classifying rail flaws through advanced image analysis techniques. We investigate four different approaches, including VGG-16 (Visual Geometry Group) with and without data augmentation and a modified VGG-16 architecture with and without data augmentation. Our results demonstrate that the modified VGG-16 architecture with data augmentation achieves the highest classification accuracy of 94.11%, surpassing the performance of the other methods. This research highlights the potential of tailored deep learning models and data augmentation techniques in achieving efficient and accurate defect detection, which helps increase this railway component reliability and improving safety improving railway safety.