[Gaussian process regression and its application in near-infrared spectroscopy analysis]

Area of Science:

• Chemometrics
• Spectroscopy
• Machine Learning

Context:

• Near-infrared (NIR) spectroscopy is crucial for analyzing agricultural products.
• Understanding complex relationships between spectral data and chemical composition is challenging.
• Accurate ingredient analysis (oil, starch, protein) in corn is vital for quality control.

Purpose:

• To apply Gaussian process (GP) regression as a chemometric method for NIR spectral analysis.
• To optimize GP models by incorporating outlier detection (Monte Carlo cross-validation) and variable selection (uninformative variable elimination).
• To evaluate the performance of GP regression for predicting corn ingredients.

Summary:

• Gaussian process (GP) regression was employed as a chemometric technique to model the relationship between NIR spectra and corn ingredients.
• Data preprocessing included outlier removal via Monte Carlo cross-validation (MCCV) and feature selection using uninformative variable elimination (UVE).
• Optimal GP regression models achieved high accuracy for predicting oil, starch, and protein content, with correlation coefficients (r) above 0.99 for calibration and 0.96 for prediction.

Impact:

• Demonstrates the effectiveness of Gaussian process (GP) regression as a powerful chemometric tool for NIR spectral analysis.
• Highlights the utility of MCCV for outlier detection and UVE for variable selection in spectral modeling.
• Shows promising results for GP algorithm in agricultural product analysis, particularly for corn ingredient quantification.