近红外光谱技术快速检测原料乳中掺杂物

采用近红外光谱技术结合化学计量学方法,对原料乳中常见的2种掺杂物——大豆分离蛋白与植脂末进行定量分析研究。先通过不同光谱预处理方法结合偏最小二乘法（PLS）建模评价不同预处理方法的效果,结果表明通过平滑处理结合多元散射校正（MSC）进行光谱预处理效果最佳,大豆分离蛋白PLS定量模型相关系数（R2）与交叉验证均方差（RMSECV）分别为0.980 9、0.127 5,植脂末PLS模型分别为0.972 2、0.130 8。随后比较了不同建模方法的效果,结果发现：采用径向基神经网络（RBF）对大豆分离蛋白的建模效果最佳,R2为0.999 4,测试集均方根误差为0.003 1;采用广义回归神经网络（GRNN）方法对植脂末建模效果最佳,R2为0.998 9,测试集均方根误差为0.004 5。因此,合理结合近红外光谱技术与化学计量学方法可快速、准确检测原料乳中大豆分离蛋白和植脂末这2种掺杂物含量。

Rapid determination of adulteration in raw milk by near-infrared spectroscopy analysis

Quantitative determination of two kinds of common adulterants in raw milk, namely soy protein isolates and creamer,was performed by combining near-infrared spectroscopy with chemometrics analysis. Firstly, the effects of different spectrum pretreatment methods were evaluated by combining different spectrum pretreatment methods with partial least squares model. The results showed that smoothing algorithm combined with multiple scattering corrections had best performance. R2 and RMSECV of PLS model for soy protein isolates were 0. 980 9 and 0. 127 5 ,and that of creamer were 0. 972 2 and 0. 130 8, respectively. Then, the performances of different modeling methods were compared. The results demonstrated that the best modeling methods for soy protein isolates and creamer were RBF and GRNN methods. R2 and RMSE test of the RBF model for soy protein isolates were 0. 999 4 and 0. 003 1, and that of the GRNN model for creamer were 0. 998 9 and 0. 004 5, respectively. Therefore, NIRS combined reasonably with chemometric methods could be used to determine the contents of soy protein isolates andcreamer in raw milk with high speed and accuracy.