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

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