采用正交投影偏最小二乘法快速无损分析乳粉蛋白质含量  

The research of fast nondestructive analysis of the protein content in milk powder based on OPLS algorithm

作　　者：邹婷婷[1] 何佳艳[1] 齐庆璇 王莹 尤梦晨 

机构地区：[1]北京工商大学北京市食品风味化学重点实验室/食品添加剂与配料北京高校工程研究中心,北京100048 [2]吉林省食品检验所,吉林长春130022

出　　处：《食品与发酵工业》2016年第4期179-182,共4页Food and Fermentation Industries

摘　　要：应用近红外光谱分析技术,建立了不同品牌不同种类不同批次的乳粉原样和混合样的蛋白质定量分析模型。采用正交投影偏最小二乘法(orthogonal partial least squares,OPLS)建立近红外光谱回归模型,并与其他预处理方法和传统偏最小二乘法(partial least squares,PLS)对比;采用交叉验证法(cross-validation)全局寻优方式获得OPLS和PLS模型的最佳参数;5个主成分建立的OPLS校正模型效果最佳,相关系数R为0.994 0,校正集交叉验证均方根RMSECV为1.09,预测集的化学值与模型预测值的相关系数R达到0.976 7,分析模型的预测误差均方根RMSEP为0.905。结果表明:OPLS回归方法在简化模型的同时提高了模型的预测泛化性能,能够快速无损建立乳粉的蛋白质近红外定量模型。With the application of near-infrared spectrum spectroscopy technique,protein quantitative analysis models for both mixed and original milk powder samples of different brands and different batches was set up. Orthogonal projection partial least squares method was adopted to establish the near-infrared spectrum regression model,and this method was compared with other pretreatment method and traditional partial least squares method; cross validation global optimization method was used to obtain the best parameters of OPLS and PLS models; OPLS correction model established by five principal components has the best effect,and the correlation coefficient R is 0. 994 0,cross-validation root mean square of calibration set is 1. 09,correlation coefficient R of chemical values of prediction set and model prediction is 0. 976 7,the error root mean square of the prediction of the analysis model is 0. 905. The result shows that OPLS regression method simplified the model and at the same time improved the generalization performance of this model,and could rapidly and nondestructively establish the near-infrared quantitative model for testing protein content in milk powder.

关 键 词：近红外光谱技术 乳粉 正交投影偏最小二乘法(OPLS) 定量分析 

分 类 号：TS252.7[轻工技术与工程—农产品加工及贮藏工程] O657.33[轻工技术与工程—食品科学与工程]