Application of a hybrid variable selection method for determination of carbohydrate content in soy milk powder using visible and near infrared spectroscopy.

Visible and near-infrared (Vis-NIR) spectroscopy was investigated to fast determine the carbohydrate content in soy milk powder. A hybrid variable selection method was proposed. In this method, a simulate annealing (SA) algorithm was first operated to search the optimal band (OB) in the wavelet packet transform (WPT) tree. The OB with 47 variables was further selected by SA (WTP-OB-SA). Finally, the number of variables was reduced from 47 to 20. The best partial least-squares prediction with a high residual predictive deviation (RPD) value of 12.2242 was obtained using these 20 variables with the correlation coefficient (r) and root-mean-square error of prediction (RMSEP) being 0.9967 and 0.1669, respectively. The results indicated that Vis-NIR spectroscopy could efficiently determine the carbohydrate content in soy milk powder. The WPT-OB-SA selection method eliminated redundant variables and improved the prediction ability.