Intelligent Concrete Surface Cracks Detection using Computer Vision, Pattern Recognition, and Artificial Neural Networks

This paper presents an intelligent concrete surface crack detection model, using machine vision, pattern recognition, artificial neural networks, and histogram of oriented gradients to feature extraction. This is an applied research category that so far, has been used in face recognition systems, and scaled conjugate gradient to training purpose multilayer neural networks, to achieve the monitoring of concrete surfaces to increase the protection that these are the innovations of this research. The main purpose of this research is to provide a reliable method for inspecting sensitive structures such as bridges using a continuous intelligent inspection system consisting of a camera and intelligent software. For this purpose, we used the Utah State University image dataset that contains over 56,000 images of cracked and non-cracked concrete bridge decks, walls, and pavements. Moreover, it includes data on cracks as narrow as 0.06 mm and a width of 25 mm with a variety of obstructions, shadows, surface roughness, scaling, edges, holes, and background debris. The ANN proposed in this study can recognize cracking images from non-cracking images. The efficiency of this network is 84.88% and it reduces network error from 0.631 to 0.199 and gradient index from 25.5 to 0.0801. This model will be useful for structural health monitoring and other sensitive surfaces such as aircraft engine shells and train moving rails. Based on the obtained results, the difference between the proposed method and other neural networks such as Convolutional Neural networks is that the former is simpler in coding and needs cheaper hardware.