作　　者： ; ; ;

机构地区： 中国科学院深圳先进技术研究院

出　　处： 《计算机与应用化学》 2011年第3期307-310,共4页

摘　　要： 利用近红外光谱技术对奶粉中含有多种掺假物的情况进行了定性判别分析。样本集除12个纯正奶粉样品外,146个掺假奶粉样品中分别加入了植脂末,天然大豆分离蛋白粉和麦芽糊精中的1～3种,掺假物的总含量范围在1.96%～35.9%之间。用Thermo Scientific Antaris MX型近红外光谱仪测量样本近红外光谱,采用非线性迭代偏最小二乘法(NIPALS)提取主成分,然后利用马氏距离进行线性判别分析,建立了1个8类判别模型。在138个样本集的交叉验证中,判别准确率达99.28%,20个测试样本的判别准确率达100%。另一方面,将3种掺假物中的1种作为未知干扰掺假物,用不含未知干扰物的样本建立了定性判别模型,然后用含有未知干扰物的样本进行验证。在植脂末作为未知干扰的情况下,判断奶粉是否掺假的准确率有100%,判断大豆分离蛋白粉和麦芽糊精是否掺入的准确度分别为78.94%和88.42%;而使用麦芽糊精作为未知干扰物时,调用模型判别奶粉是否掺假,准确率依然有100%,但判断植脂末和大豆蛋白粉掺入情况的准确率只有34.74%和32.63%。研究表明近红外光谱技术可以对奶粉中的掺假情况和掺假物种类进行快速判断;当掺假物定性判别模型遇到未知干扰时,使用该方法虽然可以对奶粉是否存在掺假进行判断,但对掺假物种类难以进行正确判别,因此建模样本集应包含尽可能多类型的掺假物。 Near Infrared Spectroscopy （NIRS） was used to discriminate the adulterate milk powder with adulterations such as vegetal creamer, bean albumen powder and maltodextrin. We got a set of 158 samples. 12 samples was pure milk powder without adulteration, the other 146 samples contained one or more adulterations above whose total contents were from 1.9% to 35.9% .The samples were divided into the modeling set of 138 samples and the validation set of 20 samples. Thermo Scientific Antaris MX FT-NIR Process Analyzer was used to get the Near-infrared spectra of those samples. The Nonlinear Iterative Partial Least Squares （NIPALS） algorithm was applied to compress the original spectrum data to 8 principal components. Then the Mahalanobis distance was calculated to make linear discriminations. Finally an 8-classification model was developed. The discrimination accuracy was 99.28% when cross validation was carried out on the modeling set. While using the validation set to validate the model, the discrimination accuracy was 100%. In order to investigate the applicability of the model while the samples contain unknown adulteration, we took one of the three adulterations as the unknown adulteration. Set the ＂samples which did not contain the unknown adulteration as calibration set, established 3 classification models which were used to discriminate the other two known adulterations. And then, used those samples that contained the unknown adulteration to validate. The discrimination accuracy of adulterate milk was still 100%. But to denote the types of the adulterations, the accuracy was decreased to 30%-90%. Results showed that NIRS could be an effective means to discriminate the adulterate milk powder and denote the types of adulterations even though there were multiple adulterations in the milks. However, if an unknown type of adulteration existed, the method could work well to discriminate whether adulterations existed, but hardly did it can denote the adulteration types. Thus, the modeling sample set should contai

关 键 词： 近红外光谱技术 奶粉 掺假

领　　域： [理学] [理学] [化学工程]