Exploring Deep Learning Methods for Accurate Bridge Crack Detection: A Comparative Study

Bridges serve as an essential part in infrastructure and must undergo consistent inspection and maintenance to ensure safety and structural soundness. One common issue that bridges can face is the development of cracks, which can compromise their stability and safety. This study proposed a crack detection system with the Kaggle dataset of bridge images with and without cracks and trained using three architecture SqueezeNet, GoogLeNet and Alex-Net architecture. The images were pre-processed to improve their quality and to standardize their dimensions. A transfer learning approach was applied to a pre-trained network to classify cracked and non-cracked images. Its performance was evaluated using metrics including accuracy, precision, recall, and F1-score. The experimental findings indicated that the AlexNet-based system achieved remarkable crack detection accuracy exceeding 99% in bridge images. Furthermore, it showcased high precision, recall, and F1-score, emphasizing its capability to effectively identify cracks while minimizing both false positives and false negatives.