Bridge Crack Detection Based on Convolutional Neural Networks

Abstract

Identifying cracks in bridges and determining the status of bridges are primarily manual labor-intensive tasks. Bridge inspection by human experts has some problems, such as the difficulty to physically inspect all portions of the bridge and the reliance on the expert knowledge of the bridge inspector as the sole source of information. Moreover, sufficient training of the human resource is required, and the overall cost of the solution is not cost effective. Data obtained from newspapers, web portals, and other sources indicates that there have been reports of many fatalities in the state of Maharashtra as a result of bridge and civil structural collapses, according to the data. Most of these incidents were reported to have occurred as a result of ignorance and lack of maintenance, which was caused by the high cost of human resources and manual inspection, as noticed in the majority of the incident reports. Using wavelet-based image characteristics in conjunction with CNN, this research provides an automatic bridge inspection method that can detect cracks in bridge photographs without the need for human intervention. In order to develop a system that can detect cracks in bridges, various bridge crack photos are collected and used as samples. A two-stage technique is used, with the first stage determining whether or not an image should be subjected to a pre-processing phase based on the image's attributes. The scanning and processing of an image should be done in accordance with the characteristics of the image. Wavelet characteristics are extracted from the image later in the second stage using a sliding window-based technique, which is described in detail below. The converted image contains the extracted features, which can be used to compare the input image with several images from a dataset, as seen below. The Convolution Neural Network Algorithm aids in the extraction of image features and the conversion of the image to grey scale. It is a type of neural network algorithm. The model is developed for the identification of bridge cracks, and it is stimulated and trained using the MATLAB programming language. It is possible to determine the crack classification of a bridge.