CNN based multi-output regression model to estimate infrastructural surface crack dimensions adopting a generalized patch size and FWHM-based width quantification

Abstract To cater the need for real-time crack monitoring of infrastructural facilities, a CNNRegression model is proposed to directly estimate the crack properties from patches. RGB crack images and their corresponding masks obtained from a public dataset are cropped into patches of 256 square-pixels that are classified with a pre-trained Deep Convolution Neural Network, the True Positives are segmented and crack properties extracted using two different methods. A statistical test has been performed for comparison and a database prepared with the more suitable method which is then fed into the neural network model to predict crack-length, crack-width and widthuncertainty directly. The proposed model has been tested on crack patches collected from different locations. Huber Loss has been used to ensure robustness of the proposed model selected from a set of 288 different variations of it. In spite of using a limited sized data set, quite satisfactory results have been achieved.