复方板蓝根颗粒中靛蓝和靛玉红含量的近红外光谱检测模型研究

摘要

[目的]提出一种用近红外光谱技术快速检测复方板蓝根颗粒中靛蓝和靛玉红含量的新方法.[方法]首先应用光谱仪获得6种复方板蓝根颗粒的光谱曲线,用主成分分析法进行聚类分析,再结合人工神经网络技术建立模型进行检测.在主成分分析的基础上,以每一个样品的前7个主成分作为神经网络的输入节点,成分类型作为神经网络的输出节点,建立一个7(输入节点)-7(隐含层节点)-2(输出节点)的3层BP人工神经网络模型.[结果]复方板蓝根颗粒中靛蓝和靛玉红2项指标人工神经网络模型预测值的平均相对误差分别为4.14%和4.72%,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.[结论]新模型可用于复方板蓝根颗粒的质量检测和生产加工过程中的质量控制.%[Objective] The aim was to put forward a new method for the detection of indigotin and indirubin contents in compound indigowoad root granule by using near infrared reflectance spectroscopy (NIRS).[Method]Firstly, through the principal component analysis method (PCA) to analyze spectroscopic curves of 6 kinds of compound indigowoad root granule which were obtained by spectrometer, then, combined with artificial neural network technology, the model was established to determine. The clustering of indigotin and indirubin content of indigowoad root granule was processed. Based on the PCA results, the first seven principal components were applied as ANN-BP input nods, and the 2 predictive indexes were applied as output nods, a testing model of artificial neural network(ANN-BP) with 7(input nods)-7(hidden layer nods)-2(output nods) was set up. [Result]The average relative error of NIRS model of indigotin and indirubin content were 4.14 % and 4.72%. The predicted value nearly was equal to HPLC value. The NIRS model had good predictability to analyze compound indigowoad root granule quality. [Conclusion]This NIRS model can be used on detecting compound indigowoad root granule quality and quality controlling of compound indigowoad root granule production processing.