A knowledge-driven approach of classifying hyperspectral images of geothermal samples

Identifying hydrothermal alteration minerals on both surface and subsurface (i.e. drill cores and cuttings) samples is essential to understand the hydrothermal processes and history of a geothermal system. With the development of infrared imaging spectroscopy, the future mineral identification analyses in geothermal exploration will likely use a lot of this new technique. The measurement collects the image of the samples as well as one infrared spectrum for every pixel of the image. As a result, one image contains a lot of infrared spectra. Thus, it is challenging to do a manual interpretation on individual spectra. The mineral identification is then expected to move towards a more autonomous way. Our research aims to test a classifier tool which is designed to mimics the knowledge and approach of spectral geology experts. We use the information on the commonly occurring minerals in geothermal systems and target their diagnostic spectral features to build the mineral classes. We combine the spectral characteristics of each of these minerals with the importance of them on deducing the hydrothermal conditions of geothermal systems for deciding the order of mineral classes. So in the case of pixel with a spectrum containing mineral mixtures, the pixel will be classified as the more important mineral class. In this paper, we would like to demonstrate how the classifier is built and what the expected outcomes look like. We aim for a classifier that can be applied reliably by non-experts working in any geothermal system.