Headspace Analysis of Whole Milk Quality Using a Surface-Enhanced Raman Spectroscopy-Active Fiber and Convolutional Neural Networks

Monitoring changes in headspace volatile compounds (VOCs) in whole milk is a well-established approach to evaluate its quality. In this study, a surface-enhanced Raman spectroscopy (SERS)-active fiber was employed to monitor headspace VOCs of whole milk during an accelerated storage condition (25 °C), and the resulting spectral data were correlated with key milk quality indicators, including pH, microbial growth, zeta potential, and electrical conductivity, using convolutional neural networks (CNNs). The results demonstrate that the SERS-active fiber effectively detected an increase in the intensity of dimethyl sulfide, a key marker of milk spoilage, during storage. Furthermore, CNN regression provided better R values (0.77, 0.79, 0.85, and 0.87) between the spectral data and quality indicators: electrical conductivity, pH, zeta potential, and microbial growth, respectively, compared to traditional chemometric (non-CNN) methods (0.69, 0.72, 0.72, and 0.68, respectively). These findings emphasize the potential of VOC analysis using the SERS-active fiber combined with CNN models for characterizing the change in milk quality during storage.