Fusion of elemental and molecular fingerprints for accurate classification of kimchi by country of origin

Abstract

The geographical origin of commercial kimchi products is a key indicator of their quality, authenticity, and economic value. In this study, we propose a spectroscopic classification method combining laser-induced breakdown spectroscopy (LIBS) and infrared (IR) spectroscopy to differentiate kimchi samples from South Korea and China. LIBS was used to obtain elemental profiles based on the emission intensities of K, Mg, Na, Ca, C, H, and O, while IR spectroscopy captured molecular features. Principal component analysis of IR spectra in the carbohydrate absorption region (1254–1018 cm⁻¹) identified the third principal component (PC3) as the most discriminative. Classification models using k-nearest neighbors (k-NN) were evaluated with leave-one-out cross-validation. Two LIBS-only models—using variable sets (i) K I (766 nm), O I (777 nm), C I (248 nm), and (ii) K I, O I, Mg II (279 nm)—achieved 94.4% accuracy. The IR-only model reached 86.4%. Fusion of LIBS and IR features, with optimized weighting for the IR variable, enhanced model performance. The best result (96.8% accuracy) was achieved by combining LIBS variables K I, O I, and C I with IR PC3. We also introduce a statistical method to predict the optimal weighting factor for fusion, reducing computational complexity by minimizing the number of neighbors in k-NN. This LIBS-IR fusion strategy provides a robust tool for verifying kimchi origin.