Prediction of bentonite water content using visible and near-infrared spectroscopy combined with partial least squares regression

Abstract

This study proposes a non-destructive method for accurately predicting the water content of bentonite using hyperspectral imaging combined with partial least squares regression (PLSR). Hyperspectral data were collected across the visible (400–700 nm) and near-infrared (1300–1600 nm) spectral ranges from bentonite samples with six controlled water content levels (0, 5, 10, 15, 20, and 25 %). Separate PLSR models were developed for the visible (VIS), near-infrared (NIR), and combined VIS + NIR spectral ranges. Among these, the VIS + NIR model demonstrated the highest predictive accuracy, achieving an R² of 0.9975 and RMSE of 0.4309 %, significantly outperforming models using individual spectral ranges. The enhanced performance of the combined model is attributed to the integration of macroscopic brightness changes captured in the VIS region and water-specific absorption features in the NIR region. This method provides a rapid and reliable approach for water content prediction, offering significant potential for quality control in bentonite buffer material production and other moisture-sensitive industrial applications.