基于主成分分析和人工神经网络的五味子质量鉴定方法研究

提出了一种采用近红外光谱技术结合人工神经网络对中药五味子质量进行鉴别的新方法.利用近红外光谱仪获得了3种不同来源地五味子合计90个样本的光谱曲线,采用主成分分析法对光谱数据进行了聚类分析,并结合人工神经网络技术建立了五味子甲素、五味子乙素和五味子醇甲三种木脂素类化合物的分析模型.主成分分析表明,前5个主成分的累积贡献率为98.75%,具有很好的聚类作用.在主成分分析的基础上,取前5个主成分的18个吸收峰作为网络的输入节点,取3项指标作为输出节点,建立了一个18(输入节点)-10(隐含层节点)-3(输出节点)的三层人工神经网络模型.五味子甲素、五味子乙素和五味子醇甲三项指标的人工神经网络模型预测值的平均相对误差分别为4.07%、2.65%和6.15%,与高效液相色谱法测定值的符合程度很高.该模型具有很好的预测能力,可用于大批量五味子的质量检测和五味子生产加工过程中的质量控制.

Study of Schisandra Chinensis Quality Discrimination Based on Principal Component Analysis and Artificial Neural Network

A new method for discriminating the quality of schisandra chinensis bωed on near infrared spectroscopy and artificial neural network is proposed. The spectral curves of 90 schisandra chinensis samples from three different sources are obtained by using a near infrared spectrometer. The principal component analysis (PCA) method is used to make cluster analysis of the spectral data and to establish the analysis models of schizandrin A, schizandrin B and schsantherin A. The PCA result shows that the first five principal components have their cumulative reliability of 98.75%. The clustering is good. On the basis of the PCA result, an ANN model with 18 input nods, 10 hidden layer nods and 3 output nods is established by using 18 absorption peaks of the first five principal components as input nods and taking 3 specifications as output nods. The ANN model predicts that the average relative error for 3 specifications of schizandrin A, schizandrin B and schsantherin A are 4.07%, 2.65%, 6.15% respectively. This is in good agreement with the prediction of HPLC. This model has an excellent prediction capability and can be used for quality checking and quality control of schisandra chinensis in a mass production process.