A 3D correlation spectroscopy framework for spectral feature decoupling: Enabling precise authentication and...

Traditional Raman spectroscopy for dairy cream fat (DCF) authentication is often severely compromised by temperature perturbations, which alter crystalline orderliness and obscure low-concentration adulteration signals. To overcome this limitation, this study establishes a novel analytical framework integrating Raman spectroscopy, three-dimensional correlation spectroscopy (3D-COS), and a custom 3D-CNN architecture.

By constructing 3D spectral data cubes under dual perturbations (temperature and concentration), 3D-COS successfully decoupled physical thermal responses from chemical structural disruptions. Spectral analysis revealed that temperature modulates the triglyceride packing conformation, altering the vibrational uniformity at 960 cm^-1, whereas adulterants explicitly disrupt the crystalline network, producing distinct variations at 960 cm^-1 and 1655 cm^-1. To thoroughly exploit these high-dimensional features, a 3D-CNN model was developed and rigorously evaluated using sample-level dataset partitioning to prevent data leakage. The proposed 3D-CNN effectively captured the spatial-concentration coupling features, outperforming PLSR, SVR, XGBoost, 1D-CNN, and 2D-CNN baselines. Ultimately, the framework achieved 100% qualitative classification accuracy and exceptional quantitative precision (R² > 0.99) for adulterants.

This study overcomes the analytical limitations caused by temperature fluctuations. By effectively decoding the highly discriminative coupled features across wavenumber space and concentration dimensions, the proposed 3D-COS and 3D-CNN framework provides a reliable, anti-interference, and mechanistically grounded technical solution for rapid dairy authenticity screening.