A feasibility study on on-line determination of rice wine composition by Vis–NIR spectroscopy and least-squares support vector machines

The feasibility of visible and near infrared (Vis–NIR) spectroscopy and least-squares support vector machines (LS-SVM) for on-line determination of rice wine composition was investigated. A circle-light fibre spectrometer system was designed to collect transreflectance spectra of rice wine samples in round brown glass bottles with the bottle sealed and the labels removed. Statistical equations were established between reference data and Vis–NIR spectra by LS-SVM. Compared to partial least squares regression (PLSR), the performance of LS-SVM was slightly better, with higher correlation coefficients for validation (rval) of 0.915, 0.888 and 0.872, and lower root mean square error of validation (RMSEP) of 0.168 (%(V V−1)), 0.146 (g L−1) and 0.033 for alcohol content, titratable acidity, and pH, respectively. Based on the results, it was concluded that the Vis–NIR spectrometer system was suitable for on-line wine quality determination, and LS-SVM was a reliable multivariate method for NIR analysis.