Multi Pixel Stochastic Approach to Mineral Samples Spectral Analysis for Geometallurgical Modeling
Published in Procemin Geomet, 2020
Recommended citation: Cristian F. Jara, Alejandro Ehrenfeld, Álvaro F. Egaña, Christian Vidal, Felipe A. Santibáñez-Leal. (2020). "Multi Pixel Stochastic Approach to Mineral Samples Spectral Analysis for Geometallurgical Modeling." Procemin Geomet. pp 1 - 17. https://gecamin.com/procemin.geomet/index.php#home
This work proposes a novel approach for analysis of spectra data for the mining industry, that uses multi pixel hyperspectral images and statistical analysis techniques to overcome the effects of ambient conditions and geological context variability to estimate critical geological and geometallurgical interest variables.
A hierarchical regression scheme is used to characterize correlation information among spectra, in order to capture statistically significant properties of different spectrum clusters for the estimation process.
A set of experiments were developed considering white reference spectra to evaluate variability and how it affects the spectra normalization process. A set of mineral samples with different geometallurgical content were used to show the proposed approach and a set of monopixel spectra were grouped to estimate their belonging to different geometallurgical measurements using the stochastic approach as well.
These experiments results emphasize that the proposed methodology improves robustness of the estimation tasks, which is considered by the research group as a relevant innovation in the use of spectral based data in the characterization of geometallurgical variables in industrial processes.