MULTI-LABEL CLASSIFICATION FOR DRILL-CORE HYPERSPECTRAL MINERAL MAPPING

We developed a deep learning model that uses hyperspectral imaging data to predict key physical rock properties, specifically density, slowness, and gamma-ray values. Our model successfully learned to translate hyperspectral information into predicted physical properties. Tests on independent data gave accurate results, demonstrating the potential of hyperspectral data for mapping physical rock properties.

Volcanic rocks are shaped by many processes, including volcanism, chemical alteration and weathering. These processes change the rock's properties, making it difficult to predict volcanic hazards or design tunnels and mines in volcanic areas. In this study, we build on earlier research to connect unique spectral signatures that can be remotely imaged using hyperspectral cameras to the density, porosity, strength, and stiffness of volcanic rocks.

We apply multi-range hyperspectral imaging on drill core material from a Kupferschiefer-type Cu–Ag deposit in Germany, mapping minerals such as iron oxides, kaolinite, sulfate, and carbonates at millimetre resolution and in a rapid, cost-efficient, and continuous manner to track hydrothermal fluid flow paths and vectors towards base metal deposits in sedimentary basins.

Novel methods for mineral exploration play an important role in future resource exploration. The methods have to be environmentally friendly, socially accepted and cost effective by integrating multidisciplinary methodologies. We investigate the potential of passive, ambient noise tomography combined with 3D airborne electromagnetics for mineral exploration in Geyer, Germany. We show that the combination of the two geophysical data sets has promising potential for future mineral exploration.