Rock Image Classification Using CNN Assisted with Pre-processed Cellular Automata-Based Grain Detected Images

The mining industry plays a pivotal role in modern society, driving economic growth and sustaining our quality of life. A core challenge within this industry is the accurate identification of rock types, traditionally categorized as igneous, sedimentary, and metamorphic. Conventional methods for this classification, relying on visual assessments, spectroscopy, and chemical analyses, are not only time consuming but also often lack consistency. To address these limitations, this study explores the integration of Convolutional Neural Networks (CNNs) into the mining sector. CNNs, renowned for their image classification capabilities, have seen widespread use in various domains but are underutilized in mining. We focus on classifying rock images with a specific emphasis on granular structures, a promising application previously demonstrated. We introduce Cellular Automata (CAs) to automate grain boundary extraction within rock images. We further investigate the potential of CA-based edge detection coupled with CNNs for rock image classification. Precisely identifying grain boundaries enhances the efficiency of classification techniques, reducing the need for exhaustive iterations. Comparative analysis of the proposed model with two other CNN architectures are shown in the paper. Our proposed model shows superior result compared to the other models. Also, efficiency of CA based model is shown compared to the CNN model without using CA.