Support vector regression for prediction of stable isotopes and trace elements using hyperspectral imaging on coffee...

Area of Science:

• Analytical Chemistry
• Geochemistry
• Agricultural Science

Background:

• Traditional geochemistry analysis is time-consuming and destructive.
• Origin determination of agricultural products like coffee is crucial for quality control and authenticity.

Purpose of the Study:

• To explore the potential of near-infrared hyperspectral imaging (HSI-NIR) for predicting stable isotope and multi-element datasets in green coffee beans.
• To assess the feasibility of using support vector regression (SVR) for this prediction task.

Main Methods:

• Utilized push-broom HSI-NIR (700-1700 nm) on green coffee beans from diverse origins.
• Analyzed five isotope ratios (δ13C, δ15N, δ18O, δ2H, δ34S) and 41 trace elements.
• Applied support vector regression (SVR) with a radial basis function kernel to correlate HSI-NIR data with geochemical markers.

Main Results:

• Achieved high prediction accuracy (R² 0.70-0.99) for three isotope ratios (δ18O, δ2H, δ34S) and eight elements (Zn, Mn, Ni, Mo, Cs, Co, Cd, La) across continental, country, and regional scales.
• All five isotope ratios were well predicted at country and regional levels.
• Identified key wavelength regions contributing to prediction models and discussed geochemical parameter correlations.

Conclusions:

• HSI-NIR is a feasible, rapid, and non-destructive method for estimating traditional geochemistry parameters in coffee beans.
• The predicted geochemical parameters are valuable origin-discriminating variables linked to environmental factors like altitude, temperature, and rainfall.