Online determination of quality parameters of dried soybean protein–lipid films (Fuzhu) by NIR spectroscopy combined with chemometrics

Abstract

Fuzhu, a protein–lipid film, is formed during the heating of soymilk. Protein, lipid and moisture contents are key quality parameters, which also affect commercial pricing in the Fuzhu industry. Near-infrared (NIR) technology was investigated for the online determination of Fuzhu quality parameters. The spectra (1000–2499 nm) of intact Fuzhu collected from four production lines during a complete production cycle were recorded on a NIR spectrometer using diffused reflectance mode. A hybrid approach combining wavelet transform with derivative calculation and multiplicative scatter correction optimally enhanced the spectral characteristic signals. Random frog (RF) and successive projections algorithm (SPA) were used for determination of key variables for PLS and MLR modeling. Comparing to the routine PLS and MLR models, the performances of RF-PLS and SPA-MLR models showed higher residual predictive deviations of 2.89 and 2.91 for protein, 3.03 and 2.97 for lipid, and 3.61 and 3.45 for moisture, respectively. An external sample set from another production line was used to assess the performances of developed RF-PLS models of protein, lipid and moisture, which yielded lower root mean squared error of prediction of 0.675, 0.554 and 0.136%, respectively, and lower absolute error value range of 1.57–1.42, 1.08–0.98 and 0.29–0.23%, respectively. Based on the predicted values obtained with RF-PLS models, a satisfactory classification of Fuzhu grades was also obtained with a total accuracy rate of 91.8%. The above results indicate that NIR coupled with chemometrics shall be a promising tool for online determination of the quality parameters of Fuzhu.