Multicomponent hyperspectral grade evaluation of ilmenite using spectral-spatial joint features

Abstract

Obtaining a comprehensive understanding of ore grade information is of significant importance for evaluating the value of ore. However, the real-time detection of multicomponent grade needs more effective online methods. This study proposes a novel approach utilizing hyperspectral imaging (HSI) to evaluate the grade information of nine major ilmenite components by integrating spectral and spatial data. Four multivariate input–output models were developed to mitigate variable interference to predict each component's grade. The results demonstrated that the backpropagation neural network (BPNN) model built from iPLS-VCPA-IRIV feature selection spectral data worked best (R_P² = 0.9935, RMSEP = 0.1364, RPD = 12.8986, and RPIQ = 21.4871, with a computational time of approximately 0.8 s). Furthermore, applying the best optimal combination algorithm for multicomponent grade inversion yielded highly accurate results, in which 97% of the component inversion residuals were less than 1. This investigation affirms that HSI enables rapid and accurate prediction and inversion of the multicomponent grade of ilmenite, thereby presenting a promising alternative to online analysis in the mineral field.