Discrimination of Milk Freshness Based on Synchronous Two-Dimensional Visible/Near-Infrared Correlation Spectroscopy Coupled with Chemometrics

Abstract

Milk is one of the preferred beverages in modern healthy diets, and its freshness is of great significance for product sales and applications. By combining the two-dimensional (2D) correlation spectroscopy technique and chemometrics, a new method based on visible/near-infrared (Vis/NIR) spectroscopy was proposed to discriminate the freshness of milk. To clarify the relationship be-tween the freshness of milk and the spectra, the changes in the physicochemical indicators of milk during storage were analyzed as well as the Vis/NIR spectra and the 2D-Vis/NIR correlation spectra. The threshold-value method, linear discriminant analysis (LDA) method, and support vector machine (SVM) method were used to construct the discriminant models of milk freshness, and the parameters of the SVM-based models were optimized by the grid search method and particle swarm optimization algorithm. The results showed that with the prolongation of storage time, the absorbance of the Vis/NIR spectra of milk gradually increased, and the intensity of autocorrelation peaks and cross peaks in synchronous 2D-Vis/NIR spectra also increased significantly. Compared with the SVM-based models using Vis/NIR spectra, the SVM-based model using 2D-Vis/NIR spectra had a >15% higher prediction accuracy. Under the same conditions, the prediction performances of the SVM-based models were better than those of the threshold-value-based or LDA-based models. In addition, the accuracy rate of the SVM-based model using the synchronous 2D-Vis/NIR autocorrelation spectra was >97%. This work indicates that the 2D-Vis/NIR correlation spectra coupled with chemometrics is a great pattern to rapidly discriminate the freshness of milk, which provides technical support for improving the evaluation system of milk quality and maintaining the safety of milk product quality.