2种判别方法对黑木耳产地分类的比较研究

Comparative Study of Two Discriminant Methods of Origin Classification of Black Fungus

采用云南、贵州、四川、福建、广西、黑龙江6个不同产地的黑木耳样本,通过ICP-MS,以镧、铈等16种稀土元素为指标,进行Fisher判别分析(FDA)和聚类分析,并比较了二者的性能优劣。结果表明,通过聚类分析,云南、贵州、四川的样品存在混聚现象,分类效果不是很好;通过Fisher判别分析,对初始案例中95.1%进行了正确判断,交叉验证分析成功率为93.3%,对36个外部验证样本进行判别回代,正确率为90.1%。Fisher判别分析方法具有突出的维数约减能力,能有效地融合训练数据的类别信息,并按照分类能力提取特征,所呈现的分类效果明显优于聚类分析方法。因此,基于稀土元素的Fisher判别分析方法可以有效地对黑木耳的产地进行分类。

The method of discriminanting classification of black fungus production based on rare-earth elements was studied by the way of using Fisher discriminant analysis （FDA） and cluster analysis, with six groups in Yunnan, Guizhou, Sichuan, Fujian, Guangxi and Heilongjiang. The study had seventeen discriminant indexes of rare-earth elements through ICP-MS. There had 95.1% in judgment though FDA. The success probolity of cross validation was 93.3%. Though back-substitution 36 external validation samples, correct rate was 90.1%. FDA could effectively reduce dimension and extract the features based on the classifying capability by caragory information in training data. So Fisher discriminant analysis based on rare-earth elements could effectively categorize the origin of black fungus.