Performance analysis of deep convolutional neural network models for railway track damage detection

Abstract Ensuring the safety and security of the Railways - one of the world’s largest transportation networks is a critical priority. Deep learning technologies present a promising avenue for enhancing track maintenance by enabling automated damage detection. The proposed work focuses on the use of Convolutional Neural Networks to identify various types of railway track defects, including cracks, corrosion, and misalignments, which are common in the Indian rail network. The study involves a comparative analysis of several Convolutional Neural Network architectures, including ResNet50, VGG, DenseNet, MobileNet, AlexNet, ConvNet, and AdaptiveSSN, using a curated and pre-processed image dataset consisting of 12,000 images. The results highlight the strong performance of ResNet50 and DenseNet, achieving accuracy rates of 95.42% and 94.35%, respectively, followed by VGG with 92.60%. This work demonstrates the effectiveness of Convolutional Neural Networks based deep learning models in railway infrastructure monitoring and offers valuable guidance for researchers aiming to refine such models for enhanced damage detection. It also outlines future research directions to improve the automation and accuracy of railway safety systems.