Short-wave near-infrared spectroscopy technique for fast determination of carbohydrate content in milk powder

Short-wave near-infrared spectroscopy (short-wave NIR) was investigated to predict carbohydrate content in milk powder. A total of seven brands of milk powder were prepared and the calibration set was composed of 234 samples, while 116 samples for the prediction set. Standard normal variate (SNV) was preformed as the spectral pretreatment. Based on the whole short-wave NIR spectra, the carbohydrate contents were well predicted. Performances of least-square support vector machine (LS-SVM) are better than those of partial least squares (PLS). Determination coefficients of LS-SVM models for prediction (Rp2) are up than 0.98, and the root mean square error of prediction (RMSEP) are less than 0.40. The loading weights of PLS and regression coefficients of PLS and LS-SVM were used to determine the sensitive wavelengths for fat content of milk powder. Optimal seven sensitive wavelengths, namely 835, 861, 897, 919, 945, 958, and 982 nm, were obtained, and the spectra at these wavelengths were used for the content determination. Rp2 of LS-SVM models are up than 0.98, and RMSEP are less than 0.40. The results indicated that short-wave NIR spectroscopy technology could be successfully applied as a fast and high precision method for the determination of carbohydrate content in milk powder.