Classification of Marble Types Using Machine Learning Techniques

Abstract

Natural stones are one of the indispensable elements of people from shelter to weapons. Among these stone types, marbles and marble-derived products are among the objects that people always prefer, from bathroom to kitchen, from garden design to small decorative home decorations. While the marbles are named according to the regions where they are extracted, their types and qualities are classified based on observation by people who are qualified as experts in this field. This classification, which is made by experts based on observation, carries risks in economic terms, increases the workload and is a difficult process with a high error rate. These processes need a fast, easy and highly accurate digital transformation. In this study, feature extraction was done by using deep learning in the species classification of marbles. The extracted features were classified using machine learning techniques. As a result of the application made with the data set consisting of 3703 marble and marble-derived natural stone images belonging to 28 different species, a classification success of 99.7% was obtained with the DenseNet deep learning model and the K-Nearest Neighbor method.