Estimation of copper concentration of rocks using hyperspectral technology

Abstract

Rock geochemical information is important for mineral exploration and provides a theoretical basis for the rapid delineation of hidden minerals. Remote sensing technology provides the possibility of rapid and large-scale extraction of geochemical information from the earth’s surface. This study analyzed the relationship between copper concentration and rock spectra by first collecting 222 rock samples, and then measuring the copper concentration of rock samples in the laboratory and reflectance spectra using an ASD FieldSpec3 portable spectrometer. It finally established quantitative relationships between the original spectra, first-order derivative spectra and second-order derivative spectra and copper concentration, respectively, using the partial least squares support vector machine method (PLS-SVM). The results show that 1) The estimation accuracy of using second-order derivatives spectra as input parameters to establish a model for estimating copper concentration is the highest, and the determined coefficient (R2) between the predicted value and real value reaches 0.54. 2) When the copper concentration is less than 80 mg/kg, the inversion model of copper concentration established using PLS-SVM obtains a good result. The R2 between the predicted copper concentration and the real copper concentration reached 0.70248. When the copper concentration is greater than 80 mg/kg, the inversion model of copper concentration established using partial least squares (PLS) obtains a good result. The R2 between the predicted copper concentration and the real copper concentration reached 0.49. The R2 between real copper concentration and copper predicted by the method of piecewise separate modeling reaches 0.816. Therefore, the method of segmental modeling has great potential to improve the accuracy of copper concentration inversion.