Extraction of Raman spectral features and rapid identification of rice origin and storage years

Origin counterfeiting and the deceptive marketing of inferior aged rice pose challenges to grain market supervision and food safety. This study proposes a method that combines Raman spectroscopy with deep learning to characterize and identify the origin and storage year of rice. Specifically, we collected nine rice samples from seven major Chinese production regions, including a three-year storage series for a representative variety. In terms of origin characterization, conventional Raman spectroscopy was utilized to analyze the vibrational modes of bulk macromolecules. Results indicated that spectral variations were primarily associated with starch microstructure, specifically reflecting differences in crystallinity and the amylopectin-to-amylose ratio driven by geographical factors and classic chemical method for starch detection was used to verify the analytical accuracy of the spectrum. Based on this, a Convolutional Neural Network (CNN) trained on these spectra achieved 98.57% accuracy in origin classification. Regarding storage year, conventional Raman spectroscopy initially revealed structural rearrangements in starch and protein induced by aging. To explore trace metabolite changes associated with rice aging, silver nanoparticle assisted SERS was used to detect an adenine signal that increased in aged rice. These results suggest that adenine may serve as indicators for future storage year discrimination. This study demonstrates that combining macro-component analysis with trace metabolite characterization provides an effective, chemically interpretable strategy for quality assessment.