基于PLS-自组织竞争神经网络近红外光谱技术对鲜乳和掺假乳的检测方法研究  

Determination of Raw Milk and Adulterated Milk Using Near Infrared Reflectance Spectroscopy Based on Partial Least Squares and Self-organizing Competitive Neural Networks

作　　者：刘波平[1,2] 荣菡[2] 邓泽元[2] 罗香[1] 

机构地区：[1]江西省分析测试中心,江西南昌330029 [2]南昌大学食品科学与技术国家重点实验室,江西南昌330047

出　　处：《分析测试学报》2008年第11期1147-1150,1156,共5页Journal of Instrumental Analysis

基　　金：教育部南昌大学食品科学重点实验室开放基金资助项目(NCU200404);长江学者和创新团队发展计划资助项目(IRT0540)

摘　　要：通过偏最小二乘法(partial least squares,PLS)与人工神经网络(artificial neural networks,ANN)联用对鲜乳和掺有植物奶油的牛乳建立识别模型。用PLS法对原始数据进行主成分压缩,采用自组织竞争神经网络建模。取前3个主成分的21个吸收峰值输入网络,学习参数为0.05,网络训练迭代次数为200,模型鉴别准确率达100%。其次建立了植物奶油掺假量的定量检测PLS模型,并采用交互校验和外部检验考察模型的可靠性,模型的校正相关系数为0.996 3,均方估计残差(RMSEC)为0.110;交互校验均方残差(RMSECV)为0.142;应用所建PLS模型对样品中植物奶油添加量进行预测,并对预测值与真值进行配对t检验,结果表明两者差异均不显著。A pattern recognition model based on partial least squares （PLS） and artificial neural networks （ANN） was established for discrimination of the raw milk and the adulterated milk blended with vegetable cream. Twenty-one absorption peak data from the first three principal componment compressed from the original data by PLS were taken as inputs of the self-organizing competitive neural network. Taking the learning parameter as 0. 05 and the training iteration number as 200, the identification accuracy of the model could be as high as 100%. The PLS prediction model for detecting the quantity of vegetable cream blended into the raw milk was thus set up and its reliability was verified by cross-validation and external-validation. The predicted correlation coefficient of the content of vegetable cream by PLS model was 0. 996 3, and the root mean square error of calibration（RMSEC） was 0. 110, while the root mean square error of cross validation（RMSECV） was 0. 142. The paired samples t-test showed that the difference between the predicted and true values was insignificant.

关 键 词：近红外光谱技术 鲜乳 掺假乳 偏最小二乘法 自组织竞争神经网络 

分 类 号：O657.3[理学—分析化学] TS252.2[理学—化学]