Hyperspectral Response Characteristics and Quantitative Prediction of Moisture Content in Different Iron Ore Samples

Abstract

The moisture content of iron ore is a key parameter affecting beneficiation efficiency, transport costs and subsequent smelting processes. Although the traditional drying and weighing method yields accurate results, it has drawbacks such as being time-consuming and causing sample damage. Accordingly, this study investigates methods for predicting the moisture content of iron ore using hyperspectral technology, using a total of nine representative iron ore samples (H1–H9) from Anshan-type magnetite in China, limonite in Australia and hematite in Brazil, covering the sub-types of limonite, magnetite and hematite. Under conditions where grade and particle size were kept consistent, different moisture content gradients were established, and spectral data in the 350–2400 nm wavelength range were acquired using a FieldSpec 4 field spectrometer. Outliers were removed using Pearson correlation analysis, followed by a comparison of 11 classical pre-processing methods and their 1–3-step cascaded combinations. The top 15 feature bands were selected using analysis of variance (ANOVA). Based on this, a mean fusion model combining partial least squares regression (PLSR) and random forest (RF) was constructed. The results indicate that moisture content is generally negatively correlated with spectral reflectance, with distinct absorption features appearing near 1440 nm and 1920 nm; the optimal pre-processing method combinations vary across different mineral samples; and the characteristic bands are primarily concentrated near wavelengths of 360 nm, 750 nm, 1140 nm, 1402 nm, 1880 nm and 2490 nm. The PLSR-RF fusion models for the H1, H2, H3, H6, H7 and H8 all achieved an R² of over 0.94 for the training set, over 0.83 for the test set, and an overall R² of over 0.91, with the H8 sample exhibiting the highest R² across the entire dataset (0.9747). Compared with existing studies, this method demonstrates a significant advantage in predictive accuracy, providing a viable solution for the rapid, non-destructive detection of moisture content in iron ore.