Content Controlled Spectral Indices for Detection of Hydrothermal Alteration Minerals Based on Machine Learning and Lasso-Logistic Regression Analysis

Document Type

Article

Publication Date

1-1-2021

Abstract

This article introduced the quantity controlled spectral indices working at mineral contents higher than 5 wt.% for detection of sericite, chlorite, and pyrophyllite, which are the representative alteration minerals of phyllic, propylitic, and advanced argillic hydrothermal alterations. The X-ray diffraction analysis revealed that the samples are mostly pure with minor content of quartz. The absorption features of target minerals showed systematic decrease in absorption depth with decrease in the mineral content, and the changes varied by mineral types. A total of 1253 target mineral spectra and 605 nontarget mineral spectra were classified by a random forest model, which achieved an overall accuracy of 97% with mineral content above 5 wt.%. Least absolute shrinkage and selection operator logistic regressions employed spectral variables of 82 bands for sericite, 132 bands for chlorite, and 84 bands for pyrophyllite with minimal spectral overlap. The overall accuracies were higher than 93.6% with R2 values ranging from 0.57 to 0.71. Because both target minerals and nontarget minerals, these indices can reliably make mineral classification. To be compatible with remote sensing images, the water-absorption bands were excluded from the indices.

Publication Source (Journal or Book title)

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Number

739

First Page

7435

Last Page

7447

This document is currently not available here.

Share

COinS