Crack Vision Effective Concrete Crack Detection with Deep Learning and Transfer Learning

The CrackVision project focuses on automating the detection of concrete cracks using advanced deep learning techniques. The backend involves training deep learning models, including VGG16, Inception_v3,  ResNet50, MobileNetV2, and a hybrid ResNet-DenseNet model, for crack detection. These models are fine-tuned through transfer learning to leverage pre-trained weights for enhanced accuracy. The system uses Grad-CAM for explainability, providing visual insight into the areas of the image that led to the prediction, aiding in decision-making. The frontend is a Flask-based web application, where users can register, log in, and upload concrete images for crack detection predictions. The system classifies images as either Crack or No Crack and visualizes the areas influencing the model's decision through Grad-CAM. This user-friendly interface helps in monitoring concrete infrastructure health, facilitating early detection of cracks to ensure the safety and durability of structures.

Keywords:

Concrete Crack Detection, Deep Learning, Transfer Learning, VGG16, Inception_v3,  ResNet50, MobileNetV2, Hybrid Model (ResNet + DenseNet), Grad-CAM, Image Classification, Flask Web Application, Predictive Maintenance, Infrastructure Monitoring, Crack Detection System