Hyperspectral Imaging for oxides and sulfides detection and classification in SWIR range by using spectral discrimination and geological considerations.

We present a new method to detect and classify ore minerals before metallurgical processing. It consists in the use of hyperspectral images in the SWIR spectral range (1000 - 2500 nm) to characterize sulfides and oxides. Usually oxides are commonly detected in the VNIR and some sulfides in the TIR. We propose that for a local geological setting mineral identification in the SWIR range considerate both the absorption bands and the shape of the reflectance curve in the whole spectral range by using spectral discrimination with appropriated classifiers. We apply this new method to data obtained from Florida Mine, Central Chile. We used the SWIR3.0 Hyperspectral Camera of SPECIM to generate hyperspectral images of drill cores that consist of pollymetalic quartz veins (Au, Ag, Zn, Pb and Cu) hosted in volcano-sedimentary sequence with mainly propilitic hydrothermal alteration. A total of 260 hyperspectral images were obtained with a pixel size of 0.7 mm. The pixels of each image were labeled manually using both spectral information and the geological mapping. the most frequently identified minerals are: quartz, chlorite, epidote, chalcopyrite, pyrite, sphalerite, galena, hematite, magnetite, clays, gypsum and calcite. We selected a representative group of hyperspectral images for training and another for testing the algorithm. We used classifiers based on Logistic Regresions wich correctly classified the 70% of pixels. Best classified minerals are the OH-bearing minerals like chlorites, clays, because of the presence of characteristic absorption bands in the SWIR range. In the case of sulfides, sphalerite is the best classified followed by galena. Pyrite and chalcopyrite sometimes are confused with quartz due to the paragenesis in the mineral deposit. In the case of oxides, hematite is better classified than magnetite.