Deep Learning-Powered Concrete Crack Classification for Improved Structural Integrity

Detection of cracks in concrete surfaces hold a great significance in ensuring the longevity and quality of the concrete structures. Cracks decrease the load bearing capacity of the structure and cracks particularly at load bearing points can lead to structural collapse. Traditional inspection methods are time consuming, less efficient, expensive which in turn depend on the expertise of the safety engineer. In recent years, deep learning techniques mainly convolutional neural networks (CNNs) provide efficient crack detection in concrete surfaces with higher accuracy. In this research, we propose a transfer learning approach using three different deep learning models namely SqueezeNet, ResNet50 and VGG19. Transfer learning is a promising solution especially when the amount of data is small. These models were trained on a dataset consisting of 12,000 images, 8400 were for training, 1800 for validation and 1800 for testing. The outcomes have demonstrated that our method is highly proficient in classifying cracks, achieving cutting-edge accuracy following the training of deep learning models. Consequently, this approach is substantiated as effective for structural inspection and maintenance.