Lightweight Deep Learning CNN Architecture Enhanced by Squeeze-and-Excitation Attention for Bridge Crack Image Analysis

Cracks that develop over time in bridges, which are an important part of transportation infrastructure, pose a structural safety risk when detected late and can lead to increased repair costs. This study aims to demonstrate the effectiveness of convolutional neural network (CNN) based deep learning architectures enhanced with attention mechanisms in the problem of classifying bridge crack images and to analyse whether these approaches offer a reliable alternative to traditional methods. In the proposed study, model training and testing processes were performed using an open-source dataset consisting of crack and crack-free images obtained from bridge surfaces. The proposed model achieved an accuracy rate of 97.36% in the predefined data splitting scenario, demonstrating higher classification performance than existing traditional methods. The results obtained demonstrate that the attention layer offers a robust solution not only in terms of high accuracy but also in terms of reliability and generalizability in critical infrastructure applications.