Rapid Discrimination of the Country Origin of Soybeans Based on FT-NIR Spectroscopy and Data Expansion

Soybeans are widely consumed in Korea, and domestic soybeans are prized over imports, resulting in a significant price difference. Therefore, screening is required to prevent fraud. Because current analytical methods are cumbersome, simple and rapid methods are required. In this study, a model for the routine discrimination of domestically grown soybeans and imported soybeans was developed using Fourier transform near-infrared spectroscopy (FT NIRS) data and partial least squares (PLS) analysis. A total of 471 soybean samples harvested between 2018 and 2020 were collected. Three PLS models using 200 or 300 samples (n) collected in 1 year and a yearly retraining model based on 2 years’ data were developed to determine the effect of data expansion on the predictive accuracy of the model. The key spectral regions were identified and optimal pretreatment selection and classification model development were carried out in OPUS 7.0. The threshold for discrimination was found to be approximately ± 40 the reference value (Korean 100, foreign 1) based on the predicted NIRS value distribution. The sensitivity, selectivity, and efficiency of the PLS models were similar even as the database size increased, although the prediction accuracy increased. The 2018 (n = 300) model achieved 98.3% and 91% prediction rates for the 2019 and 2020 models, respectively, indicating robustness. However, the 2-year combined model showed the best prediction rate of 95.9%. Thus, the developed method can distinguish Korean and foreign soybeans and does not require complicated pretreatment, suggesting its suitability to prevent food fraud.