The analysis and interpretation of monitoring data from sensors installed on railway bridges under train loading are crucial for ensuring the effectiveness of structural health monitoring (SHM) systems. Recently, artificial intelligence (AI)-based algorithms have shown promise in overcoming the limitations of traditional approaches on real-time monitoring. Among AI algorithms, convolutional neural networks (CNNs) applied to continuous wavelet transforms (CWT) have been recently employed for classifying the location and severity of damage. This study presents a new data-driven approach based on the combination of CWT and CNN to detect and localize damage in railway bridges by using data obtained from sensors placed on a bridge. The proposed vibration-based damage identification and classification system is tested on a numerical case study considering various damage scenarios, including different severities and locations of cracks. Parametric analyses are carried out to examine the effects of variations in train velocity, train mass, and the natural frequencies of the structure. The results demonstrate the effectiveness of the methodology in accurately detecting and localizing damage under different dynamic loading conditions.