T164 Using Geophysics to Evaluate and Explore for Mineral, Energy, and Groundwater Resources I

Wednesday, 28 October 2020
5:30 PM - 8:00 PM
Meeting Rooms

INCREASING THE VALUE OF HYPERSPECTRAL DATA USING ADVANCED MACHINE LEARNING TECHNIQUES

Paper number: T164-9
Program: Technical Sessions
Time: 7:30 PM – 7:45 PM

Authors

  • Katharina Pfaff Presenting Author
    Colorado School of Mines, Golden, CO
  • Amit Rotem
    Center for Advanced Subsurface Earth Resource Models (CASERM), Golden, CO
  • Alexander Vidal
    Center for Advanced Subsurface Earth Resource Models (CASERM), Golden, CO
  • Erik R. Tharalson
    Center for Advanced Subsurface Earth Resource Models, Golden, CO
  • Thomas Monecke
    Center for Advanced Subsurface Earth Resource Models (CASERM), Golden, CO
  • Luis Tenorio
    Center for Advanced Subsurface Earth Resource Models (CASERM), Golden, CO

Abstract

Knowledge of the deposit mineralogy and physical and mechanical properties of rock units is critical at many stages of project development from early exploration to mining and remediation. Hyperspectral core scanning is currently the method of choice to determine the mineralogy of ore deposits and their host rocks. Common hyperspectral sensors cover the visible to near-infrared (VNIR) and short-wave infrared (SWIR) regions of the electromagnetic spectrum. However, within the SWIR wavelength range, traditional methods of spectrum matching are inefficient to correctly identify and quantify common minerals such as feldspar, quartz, oxides, and sulfides. The aim of this project is to automate mineral identification in a fast, accurate, and efficient way by finding functional relations between hyperspectral and quantitative automated mineralogy data using advanced machine learning techniques.

We have successfully applied deep learning (CNN), support vector machines and lasso/total variation methods to predict the most abundant mineral and individual proportions of each mineral. The obtained accuracies were found to be between 70% and 90%. The confusion matrix, which compares the true mineral assignment to the predicted mineral assignment, was predominantly diagonal, indicating correct predictions of individual mineral assignments. Predictions of the modal abundances of minerals were around 95% and their spatial distribution was close to the original.

Initial upscaling tests and masking applied to drill core have shown very promising results. Thus, supervised machine learning seems to be an efficient and accurate tool in mineral identification and classification.

Geological Society of America Abstracts with Programs. Vol. 52, No. 6

doi: 10.1130/abs/2020AM-357749

© Copyright 2020 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.