A proposed concept for classifying uniaxial compressive strength (UCS) from SWIR hyperspectral data

With the development of lower-cost and portable spectral imagers and spectral radiometers, the question arises: Can hyperspectral image data be used to estimate the Unconfined Compressive Strength (UCS) of rock? Reflectance, emissivity, absorption, and transmission are fundamental properties of rock and minerals. This study focuses on correlating data from non-destructive hyperspectral images and destructive test methods.

Hyperspectral images of 32 altered granite samples were acquired in the Shortwave Infrared (SWIR). The reflectance from the 1000 to 2500 nm range of core samples was analyzed. The primary objective of this study is to identify key spectral features that correlate with rock strength and classify samples into ISRM strength categories for weak, moderately strong, and strong rock. The methodology encompasses data preprocessing, feature extraction based on deviations from the mean spectral response, and statistical analysis to identify significant spectral components. The k-Nearest Neighbor (kNN) classifier demonstrated reliable performance for moderately strong and strong rock categories, achieving an overall accuracy of 90 %. This paper outlines the experimental procedure, machine learning analysis methods, and a recommended path forward for further developing this technique. The ultimate goal is to develop additional methods for quantifying UCS from hyperspectral images of both surface and drill core data, utilizing International Society of Rock Mechanics (ISRM) classification guidelines.