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

Area of Science:

• Mineralogy and Materials Science
• Geochemistry
• Spectroscopy

Background:

• Accurate ore grade assessment is crucial for mineral resource valuation.
• Existing online methods for real-time multicomponent ore grade detection are insufficient.
• Ilmenite ore analysis requires effective methods for evaluating multiple component grades simultaneously.

Purpose of the Study:

• To develop and validate a novel hyperspectral imaging (HSI) approach for evaluating ilmenite's nine major component grades.
• To integrate spectral and spatial data for enhanced accuracy in ore grade prediction.
• To establish effective multivariate models for mitigating variable interference in multicomponent grade analysis.

Main Methods:

• Utilized hyperspectral imaging (HSI) to capture spectral and spatial data of ilmenite samples.
• Developed four multivariate input-output models for predicting individual component grades.
• Employed iPLS-VCPA-IRIV feature selection for spectral data preprocessing.
• Applied a backpropagation neural network (BPNN) model for component grade inversion.

Main Results:

• The BPNN model, utilizing iPLS-VCPA-IRIV selected spectral data, achieved excellent predictive performance (R_P² = 0.9935, RMSEP = 0.1364, RPD = 12.8986, RPIQ = 21.4871).
• The optimal model demonstrated a rapid computational time of approximately 0.8 seconds.
• Multicomponent grade inversion using the optimal algorithm resulted in high accuracy, with 97% of component inversion residuals below 1.

Conclusions:

• Hyperspectral imaging (HSI) provides a rapid and accurate method for predicting and inverting multicomponent grades in ilmenite.
• The integrated spectral and spatial data analysis effectively overcomes variable interference in complex ore matrices.
• This HSI-based approach presents a promising alternative for online analysis in the mineral industry, enhancing efficiency and accuracy.