Exploitation
Pouya Nobahar; Yashar Pourrahimian; Roohollah Shirani Faradonbeh; Fereydoun Mollaei Koshki
Abstract
Mineral reserve evaluation and ore type detection using data from exploratory boreholes are critical in mine design and extraction. However, preparing core samples and conducting chemical and physical tests is a time-consuming and costly procedure, slowing down the modeling process. This paper presents ...
Read More
Mineral reserve evaluation and ore type detection using data from exploratory boreholes are critical in mine design and extraction. However, preparing core samples and conducting chemical and physical tests is a time-consuming and costly procedure, slowing down the modeling process. This paper presents a novel Deep Learning (DL)-based model to recognize the types of kaolinite samples. For this purpose, a dataset containing the images of drilled cores and their types determined from conventional chemical and physical analyses was used. Eight Convolutional Neural Network (CNN) topologies based on individual features were developed, named A, B, C, D, E, F, G, and H. Six of the eight proposed CNN topologies described above had accuracy below 80%, whereas two of them, model A and H, had higher accuracy than other topologies. Due to their similarity in results, both of them analyzed deeply. Model A was more efficient, with 90% accuracy, than model B, with 84% accuracy. Furthermore, the class detection performance of model A was further evaluated using different indices, including precision, recall, and F1-score, which resulted in values of 92%, 92%, and 90%, respectively, which are acceptable accuracies to identify the type of samples when using this approach on six different types of kaolinite.
