Concrete and Pavement Cracks Detection Leveraging Vision Transformer (ViT) Model

Automated identification of cracks in concrete and pavement is gaining popularity today. Numerous ways have been suggested and put into practice in recent times. The primary goal is to mechanize the process with enhanced precision and efficiency. This study uses a Vision Transformer model to detect surface cracks. The primary goal is to assess the efficacy of the Vision Transformer (ViT) model and Convolutional Neural Network (CNN) in detecting concrete and pavement cracks. The model is trained and evaluated using a publicly available Surface Cracks Dataset. The collection comprises 30,000 photos, encompassing “positive” and “negative” images depicting cracked and noncracked concrete surfaces. The images are in JPEG format and have dimensions of 227 by 227 pixels in RGB format. The CNN and ViT models undergo training and validation processes, and their performance is assessed using test sets. The performance of the ViT Model is compared to a baseline CNN model. The ViT model surpasses the CNN model in accurately detecting pavement and concrete cracks, with higher accuracy rates of 99.00% and 98.33%, respectively. The results demonstrate the enhanced dependability and efficiency of the ViT model in crack detection tasks.