Content area
Quantitative mapping of minerals in rock thin sections delivers data on mineral abundance, size, and spatial arrangement that are useful for many geoscience and engineering disciplines. Although automated methods for mapping mineralogy exist, these are often expensive, associated with proprietary software, or require programming skills, which limits their usage. Here we present a free, open-source method for automated mineralogy mapping from energy-dispersive spectroscopy (EDS) scans of rock thin sections. This method uses a random forest (RF) machine-learning image classification algorithm within the QGIS geographic information system and Orfeo ToolBox, which are both free and open-source. To demonstrate the utility of this method, we apply it to 14 rock thin sections from the well-studied Rio Blanco tonalite lithology of Puerto Rico. Measurements of mineral abundance inferred from our method compare favorably to previous measurements of mineral abundance inferred from X-ray diffraction and point counts on thin sections. The model-generated mineral maps agree with independent, manually delineated mineral maps at a mean rate of 95 %, with accuracies as high as 96 % for the most abundant mineral (plagioclase) and as low as 72 % for the least abundant mineral (apatite) in these samples. We show that the default random forest hyperparameters (i.e., tuneable settings that control behavior) in Orfeo ToolBox yielded high accuracy in the model-generated mineral maps, and we demonstrate how users can determine the sensitivity of the mineral maps to hyperparameter values and input features. These results show that this method can be used to generate accurate maps of major minerals in rock thin sections using entirely free and open-source applications.
Details
Rock;
Apatite;
Lithology;
Mapping;
X-ray diffraction;
Automation;
X-ray spectroscopy;
Grain size;
Geographic information systems;
Machine learning;
Scanning electron microscopy;
Mineralogy;
Abundance;
Classification;
Spectroscopy;
Image classification;
Plagioclase;
Methods;
Algorithms;
Geographical information systems;
Software;
Information systems;
Deep learning;
Maps;
Rocks;
Open source software;
Spectrum analysis;
Electron microscopes;
X rays;
Decision trees;
Remote sensing
1 Geoscience, University of Wisconsin-Madison, Madison, Wisconsin, United States
2 Civil and Environmental Engineering, Cornell University, Ithaca, New York, United States
3 Hopkins Extreme Materials Institute, John Hopkins University, Baltimore, Maryland, United States
4 Key Laboratory of Geotechnical and Underground Engineering of Ministry of Education, Tongji University, Shanghai, China; Department of Geotechnical Engineering, Tongji University, Shanghai, China
5 independent researcher