Geometallurgical estimation from hyperspectral images and topic modelling
Date:
Presented at Procemin Geomet 2022, the premier mineral processing conference in Latin America. Formalizes mineral sample characterization as a topic modelling task using Latent Dirichlet Allocation (LDA) on hyperspectral pixels. The analogy: minerals are “topics”, spectral signatures are “words”, each pixel is a “document”. Provides experimental evidence on SWIR and VNIR wavelengths. Joint work with Alejandro Ehrenfeld, Felipe Garrido, Felipe Navarro, and Alvaro Egana.