Grade estimation in realgar-orpiment batch flotation systems by froth image analysis and kernel methods

Abstract

In this investigation, mixtures of realgar and orpiment particles were floated in a laboratory batch flotation systems and multivariate image analysis was used to estimate the arsenic content in the froths. The analysis was based on the froth colour, as well as extraction of three groups of textural features, namely those based on grey level co-occurrence matrices, wavelets and local binary patterns. Collectively, these features provided better information on the arsenic content of the froths than any one of the individual groups of features. Partial least squares models could explain approximately 78% of the variance in the arsenic by using all the features simultaneously. The colour content, particularly the green component in the red-green-blue (RGB) features, provided almost as much information as the texture-base features.