摘要
研究近红外光谱技术检测乳粉中的蛋白质和脂肪含量。挑选不同品牌不同规格的乳粉共120种,研究了光谱的预处理、奇异值的剔除、主成分数的确定、波数范围的优选等问题。以国标方法测定的结果做参比,采用偏最小二乘法建立乳粉中蛋白质和脂肪的定量检测模型,通过外部检验和交互验证两种方式考察近红外数学模型的可靠性。并利用建立的模型对未知样品进行检测。结果表明:导数处理和矢量归一化处理能够使光谱信息更加突出,蛋白质定标模型的R^2为0.952 5,RMSECV为0.984,RMSEP为0.995,RPD为4.59,而脂肪定标模型的R^2为0.973 7,RMSECV为1.27,RMSEP为1.31,RPD为6.17,两个模型均具有良好的分析精度。定标模型测定未知乳粉样品的蛋白质的相对误差均小于1.9%,脂肪的相对误差均小于1.50%,模型预测性良好。
The contents of protein and fat measured by near infrared spectroscopy were researched in this paper.One hundred and twenty different brand and Specification milk powder samples were choosed.Those problems such as preprocessing of Spectrum,eliminating singular value,determining of the principal component numbers,optimizing of wave number range were researed.The results of national standard methods were referenced.Quantitative detection models of protein and fat were established by partial least squares.The reliability of Near infrared spectroscopy mathematical model were inspected by the two ways of external inspection and interactive authentication. The unknown samples were detected by the models established in this research. Results indicated that the R^2,RMSECV,RMSEP,RPD of protein calibration models were 0.952 5,0.984,0.995,4.59 respectively.The R^2,RMSECV,RMSEP,RPD of fat calibration model swere 0.973 7,1.27,1.31,6.17 respectively.Both of the two models have good analytical precision. The relative errors of protein and fat measured by calibration model were Less than 1.9 % and 1.50 % respectively. The predictability of models were good.
作者
杨福臣
孙兆远
孙芝杨
YANG Fu-chen SUN Zhao-yuan SUN Zhi-yang(Jiangsu Food & Pharmaceutical Sicence College, Huai'an 223003, Jiangsu, Chin)
出处
《食品研究与开发》
CAS
北大核心
2017年第2期169-173,共5页
Food Research and Development
基金
淮安市应用研究与科技攻关计划(HAS2014024-4)
关键词
乳粉
蛋白质
脂肪
近红外光谱
模型
milk powder
protein
fat
Near Infrared Spectrum
model