Automatic detection of concrete cracks from images using Adam-SqueezeNet deep learning model

Cracks in the concrete surfaces are typically clear warning signs of a potential threat to the integrity and serviceability of the structure. The techniques based on image processing can effectively detect cracks in digital images. These techniques, however, are generally susceptible to user-driven heuristic thresholds and irrelevant distractors. Inspired by the recent success of artificial intelligence, a deep learning-based automated crack detection system named CrackSN was presented. This proposed deep learning model, built on the Adam-SqueezeNet architecture, automatically learns the discriminative feature directly from the labeled and augmented patches. An image dataset of concrete surfaces is collected by smartphone and carefully prepared in order to develop and train the CrackSN system. The hyper -parameters of SqueezeNet are tuned with Adam optimization through the training and validation procedures. The fine-tuned CrackSN system outperforms state-of-the-art models in recent literature by correctly classifying 97.3% of the cracked patches in the image dataset. The success of CrackSN, demonstrated by its light network design and outstanding performance, provides a crucial step toward automated damage inspection and health evaluation for infrastructure.