中文摘要：

为了有效地提取牛奶中微量的掺杂物特征信息,提出了基于二维相关近红外光谱多维主成分分析（MPCA）和最小二乘支持向量机（LS-SVM）判别牛奶掺杂物的方法.首先,采集纯牛奶、掺杂尿素牛奶和掺杂三聚氰胺牛奶的一维近红外谱,并对其进行相关计算,构建各样品的二维相关近红外谱.然后,采用多维主成分分析法分析二维相关谱矩阵,压缩数据,提取相关谱的得分矩阵.最后,将提取的得分矩阵输入最小二乘支持向量机,分别建立掺杂尿素牛奶、掺杂三聚氰胺牛奶及两种掺杂牛奶与纯牛奶的LS-SVM判别模型.用所建模型对测试集未知样品进行了判别,结果显示其判别正确率分别为92.3％,96.2％,92.3％.研究结果表明：所提出的方法不仅有效提取了牛奶中掺杂物的特征信息,而且缩短了建模所需时间,取得了较好的判别效果.

英文摘要：

T models for spectra com o extract effectively characteristic information of adulterants in milk, the classification adulterated milk were established using two-dimensional （2D） correlation near-infrared bining a Multi-way Principal Component Analysis （MPCA） with Least Square Support Vector Machines（LS-SVM）. First, one-dimensional near-infrared spectra of pure milk and adulterated milk samples were collected and the synchronous 2D correlation spectra of all samples were calculated. Then, the MPCA was used to reduce dimension by extracting score matrix of 2D correlation data set. Finally, LS-SVM models for urea-tainted milk, melamine-tainted milk, and the above two kinds of adulterated milk were constructed by using score matrix extracted from 2D correlation spectra as the input variables. These models were used to discriminate the known samples in the test set and the results show that the classification accuracy rates of unknown samples are 92.3%, 96.2%, 92.3%, respectively. It demonstrates that the proposed method not only extracts effectively feature information of adulterants in milk, but also reduces the input dimension of LS-SVM and computational time. It realizes a better classification of adulterated milk and pure milk.