ScholarWorks@경상국립대학교

초록

Rapidly accurate identification of meat species and storage conditions is critical for consumer protection, quality assurance, and fraud prevention in the meat supply chain. This study demonstrates that near-infrared (NIR) spectroscopy can serve as a rapid, non-destructive tool for multi-class meat authentication, capturing species-specific and storage-dependent variations that are traditionally assessed through time-consuming, costly, and expertise-intensive physicochemical, oxidative, chemical, and bioactive analyses. A total of 1200 meat samples from beef, chevon, and chicken were scanned across 700–1100 nm, generating 12,000 NIR spectra. Physicochemical and oxidative parameters, along with chemical composition and major bioactive compounds, including fatty acids, volatile compounds, and heterocyclic aromatic amines, were measured to validate the discriminatory power of NIR. Frozen meat exhibited higher pH, lipid oxidation, shear force, and cooking loss, accompanied by decreased redness. Furthermore, species-specific chemical and bioactive profiles provided additional confirmation of classification potential. Dimensionality reduction using Principal Component Analysis (PCA) and nonlinear embedding revealed clear separability of fresh and frozen samples across species. Machine learning models achieved high accuracy, with Logistic Regression and Neural Networks reaching the best classification. Chevon (Fresh) remained the most challenging class due to spectral and biochemical overlaps. Learning curve analyses indicated robust generalization for most models, with ensemble and neural network approaches benefiting from larger datasets. Decision boundary visualization highlighted contrasts between linear and nonlinear classifiers, as well as the smoothing effects of ensemble averaging. Overall, the integration of NIR spectroscopy with multi-type reference markers provides an efficient, accurate, and non-destructive approach for simultaneous meat authentication.

키워드