Effective railway infrastructure maintenance is vital for ensuring the safety and reliability of transportation systems. This paper investigates the efficacy of Convolutional Neural Networks (CNNs) in detecting railway rail defects without and with data preprocessing. Utilizing a database comprising 556 images of both defective and non-defective rails, we deployed two CNN configurations: (a) a two-layer CNN; and (b) a more complex five-layer CNN. Our study encompasses the evaluation of these networks before and after data balancing and implementation of different augmentation techniques. The results consistently demonstrated superior performance from the simpler two-layer CNN across all scenarios. This observation underscores the significance of network architecture over complexity, highlighting the two-layer CNN as an effective and efficient solution for rail defect detection. The model exhibits impressive performance metrics, with an accuracy of 91.58% and an F1 score of 83.49%. The findings provide valuable insights for optimizing defect detection systems in railway maintenance applications.

Railway Safety, Rail Defects, Convolutional Neural Network (CNN), Image Processing, Data Balancing, Pre-processing, Augmentation Techniques