Identification of Rock Images in Mining Industry: An Application of Deep Learning Technique

Abstract

A major part of economic upswing of a country depends on the mining industry. An exhaustive research is being held worldwide to make mining easy and cost effective. The present study deals with different types of rock identification using deep learning approach. Rock identification is considered to be an effective mining of minerals. In general, rocks are identified by skilled and expert professionals who can recognize the rocks by visual identification, spectroscopic methods, chemical analysis, etc. However, these methods are costly and time consuming. In this work, we propose a convolutional neural network (CNN) architecture to identify different types of rocks like igneous, sedimentary and metamorphic. Moreover, we classify 8 subtypes of the above rocks, such as basalt and granite (Igneous rocks); coal, limestone, sandstone and shale (sedimentary rocks); marble and quartzite (metamorphic rocks). The dataset of rock images, collected from different parts of India, are considered in the study. The proposed CNN is compared with an already established CNN—VGG16 and another established CNN model. The accuracy of our proposed model is higher compared to the other two models in case of both 3-class and 8-class classifications. Hence, the results show superiority for the proposed architecture.