[Inversion of True Protein Content in Milk Based on Hyperspectral Data]

Area of Science:

• Food Science
• Analytical Chemistry
• Spectroscopy

Background:

• Consumer concern for milk quality necessitates rapid and accurate detection methods.
• Traditional milk analysis techniques are insufficient for modern production demands.
• Visible/Near-Infrared (VIS/NIR) reflectance spectroscopy offers a potential non-destructive alternative.

Purpose of the Study:

• To develop a rapid quantitative detection method for true protein in pure milk.
• To establish and optimize a regression model using VIS/NIR spectroscopy.
• To compare the performance of Principle Component Regression (PCR) and Least Squares Support Vector Machine (LS-SVM) models.

Main Methods:

• Collected VIS/NIR spectral data (350-2500 nm) and protein content data from pure milk samples.
• Applied spectral preprocessing techniques, including multiple scatter correction and second derivative.
• Utilized band selection methods to identify optimal feature bands.
• Developed and compared PCR and LS-SVM regression models for protein content prediction.

Main Results:

• The combination of multiple scatter correction and second derivative preprocessing yielded optimal spectral data.
• Variable optimization improved model accuracy and reduced modeling time compared to using the whole spectrum.
• The LS-SVM model (R(P)²=0.9580, RMSEP=0.0482) outperformed the PCR model (R(P)²=0.9522, RMSEP=0.0487) in predicting protein content.

Conclusions:

• VIS/NIR reflectance spectroscopy, coupled with appropriate preprocessing and modeling (LS-SVM), provides an accurate and rapid method for non-destructive true protein detection in milk.
• This approach can significantly enhance milk quality control in production.
• The developed method offers a viable alternative to traditional, time-consuming protein analysis techniques.