Automatic near-infrared hyperspectral image analysis of copper concentrates

Abstract

The reflection of light on mineral samples have been widely used to obtain information concerning their composition. In particular, visible and near-infrared reflectance spectrum have offered an inexpensive way to obtain information about their mineralogical composition. In this work, near-infrared hyperspectral reflective images of several mineral samples are obtained and analyzed. The average reflective spectrum of Chalcopyrite (CuFeS2), Pyrite (FeS2), Chalcocite (Cu2S), Covellite (CuS), and Slag (FeO-SiO2) packed into pellets were obtained using a near-infrared hyperspectral camera. In order to analyze copper concentrates, a K-Nearest Neighbor classifier was trained to identify its main components. A 10 fold cross validation approach was used to certify the validity of the classifier. The trained classifier provided the mineralogical spatial distribution of the different components in a concentrate sample. An automatic system controlling all the acquisition and image processing stages provides analysis of the concentrate samples. Further work is underway to include additional minerals and to improve implementation issues such as signal filtering. This is the first step towards the design of a low cost system to provide relevant information about the concentrates feeding copper smelters.