荧光光谱法结合Fisher判别分析在西洋参鉴别中的应用

旨在建立可靠的Fisher判别模型，以实现西洋参及其常见伪品饮片的快速、客观、准确鉴别，采用自组的凝视式光谱成像仪，对90份不同市售来源的中药材饮片(西洋参、人参、桔梗各30份)进行了荧光光谱成像实验，波长范围为400~720 nm，成像间隔为5nm。采用标准正态变量(SNV)变换对原的光谱数据进行预处理，以减少光谱数据中的噪声干扰。比较了主成分分析(PCA)与逐步判别分析(SDA)的原理特点及对模型的优化效果，联合这两种分析方法，首先，应用PCA对预处理后的光谱数据进行处理，使光谱数据中的主要信息集中分布在前面的主成分中，然后应用SDA从65个主成分中筛选出判别能力较强的12个主成分建立Fisher判别模型。由所建模型的两个判别函数作样品得分散点图，各类样品在图中表现出良好的聚类现象。以待判样品点与各种类中心点之间的欧氏距离作为依据，得出模型的准确判别结果。结果显示，所建Fisher判别模型在训练集和预测集中的判别正确率分别为98.33%和 96.67%，具有较高的可信度与准确度，因此，荧光光谱法结合Fisher判别分析可用于快速鉴别西洋参及其伪品饮片。

Application of Fluorescence Spectrometry Combined with Fisher Discriminant Analysis in Radix Panacis Quinquefolii Identification

The purpose of this paper was to establish a rel iable Fisher discriminant model which was able to recognize the decoction pieces of radix panacis quinquefolii and its' common counterfeits rapidly,objectivel y and accurately.The fluorescence spectra of 90 samples (decoction pieces of ra dix panacis quinquefolii,radix ginseng and platycodon grandiflorum each 30 copi es) that from different source were detected by a self-build staring spectral i maging instrument in this study.The experimental parameters included spectral w avelengths range from 400 to 720 nm,with the interval of 5 nm.Standard normal variate (SNV) transformation was used for spectral pretreatment,to reduce the noise information in original spectral data.According to the principle features and optimizing effect of principal component analysis (PCA) and stepwise discri minant analysis (SDA),PCA in combination with PCA was needed.At first,SNV spe ctral data was processed with PCA to obtain main information of spectrum distrib uted in the first few principal components.Then 12 principal components that wi th strong discriminant ability were selected from 65 principal components,and u sed for established the Fisher discriminant model.All kind of samples showed up a good clustering phenomenon in the scatter diagram,which plotted based on sam ples scores in two discriminant functions.In order to obtain an accurate discri minant result of the model,the Euclidean distance between the central of each s pecies and the samples that under discriminate was calculated and as the gist.T he result showed that the discriminant accuracy of the Fisher discriminant model in training set and prediction set was 98.33％ and 96.67％ respectively,demons trating that the superior reliability and accuracy existed in the model.Therefo re,fluorescence spectroscopy combined with Fisher discriminant analysis could b e applied to the rapid identification between the decoction pieces of radix pana cis quinquefolii and its' counterfeits.