[Multicomponent quantitative analysis using near infrared spectroscopy by building PLS-GRNN model].

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Context:

• Accurate quantification of chemical components in feedstuffs is crucial for quality control and nutritional assessment.
• Traditional methods for analyzing feed composition can be time-consuming and labor-intensive.
• Near Infrared Spectroscopy (NIRS) offers a non-destructive and rapid alternative for chemical analysis.

Purpose:

• To develop and validate a Partial Least Squares-Generalized Regression Neural Networks (PLS-GRNN) model for multi-component quantitative analysis of feedstuffs using NIRS data.
• To assess the accuracy, recurrence, and predictive capability of the developed PLS-GRNN model for determining chlorine, fiber, and fat content.
• To optimize model parameters, such as the smoothing factor, for improved prediction accuracy.

Summary:

• A PLS-GRNN model was constructed using NIRS data from 45 feedstuff samples to predict chlorine, fiber, and fat content.
• The model utilized principal components from PLS and original spectral data as inputs, achieving high predictive correlation coefficients (e.g., 0.984 for chlorine) and low standard errors of the estimate.
• A smoothing factor of 0.1 was identified as optimal, yielding the lowest prediction error compared to other tested values.

Impact:

• The study demonstrates that the PLS-GRNN approach applied to NIRS is an effective and rapid method for quantitative analysis of chlorine, fiber, and fat in feedstuff powder.
• The findings suggest that this methodology can be extended for the quantitative analysis of other components and samples, offering a valuable tool for the feed industry.
• The research addresses challenges related to high prediction errors for samples with lower component concentrations.