A study on image-based automatic building materials classification using deep learning

Abstract

Currently, in the reverse-design process, the method of obtaining structural and morphological information of a built structure using laser scanning is commonly used, but it is difficult to acquire information of the types or properties of building material. In this study, we design a deep learning model that automatically classifies material types through images of building materials using image-based deep learning techniques and evaluates the accuracy of them. The deep learning model of this study satisfies an accuracy of 66.6% for 23 classes, but among them, 19 classes excluding some classes satisfied an average accuracy of 89.6%.