Investigation of compositional variations in chromitite ore with imaging Laser Induced Breakdown Spectroscopy and Spectral Angle Mapper classification algorithm

Abstract

This paper focusses on the applicability of the Spectral Angle Mapper (SAM) algorithm for supervised classification of imaging Laser Induced Breakdown Spectroscopy (LIBS) data. Our main objective is to investigate variations in the chemical/mineralogical composition of complex ore from the sub-millimetre to the metre scale, which may offer novel and barely investigated interpretation opportunities for exploration purposes. This research is based on coarse chromitite ore from Merensky Reef, represented by a drill core and a small section through the upper chromitite layer. Detailed LIBS-based imaging measurements were accompanied by space-resolved reference measurements based on SEM/MLA and EDXRF, as well as bulk chemical analyses for multiple core slices. The SAM algorithm was applied for classification of LIBS hyperspectral images with respect to differences in mineral chemistry. Our investigations focused on the pre-processing of LIBS spectra prior to SAM classification, on spectral library development, as well as on the validation of the classified data. The SAM classification algorithm, which is solely based on ratios between spectral intensities, was found insensitive to normal shot-to-shot plasma variations and to chemically induced matrix effects. However, the algorithm may become inaccurate at low signal to noise ratios, at the border between different mineral grains (mixed spectra), or when classifying chemically similar phases such as pyrite and pyrrhotite. The extent of mixed spectra depends both on the size of the mineral grains as well as on the spot size of the LIBS laser. The SAM algorithm was successfully applied for classification of several base metal sulphides, rock-forming minerals, accessory minerals, as well as several mixed phases representing the main borders between different mineral grains. The resulting classified LIBS image shows the spatial distribution of the different phases, which is in very good agreement with the space-resolved reference measurements based on EDXRF and SEM/MLA. The results also highlight the extremely heterogeneous distribution of e.g. the sulphide phases in the investigated core piece. The applicability of the LIBS-SAM classification image for estimating metal concentrations based on point counting has been explored for Cu, Ni, S, and Cr. We conclude that this approach, when applied on sufficiently large surfaces, enables semi-quantitative data analysis, as well as the possible detection of trace elements (e.g. Pt, Pd) that occur in very small nuggets.