Abstract: A method of fast determining active components of Chinese herb extracts by near infrared (NIR) spectroscopy, which can be used for on-line analysis of percolation extracting process, is proposed and applied to percolation extracting panax notoginseng herb as an example. Reference measurements of ginsenoside Rg1, Rb1, and Rd were performed by HPLC, and panax notoginseng saponins (PNS) were determined according to the colorimetric assays. The calibration model between NIR spectra and reference values was set up. Radial basis function neural networks(RBFNN) and partial least squares regression(PLSR) were studied comparatively to establish the calibration model.By using RBFNN, RMSECVs for Rg1, Rb1, Rd, PNS were 1.120, 1.230, 0.267, 4.749 and RMSEPs were 0.677, 0.969, 0.155, 8.065, respectively. The results show that the present method is convenient, accurate and non-destructive, and can be applied for process analysis and quality control of Chinese medicine manufacturing process.

Key words:

近红外光谱, 径向基人工神经网络, 中药, 渗漉提取

摘要： 研究提出一种快速测定中药渗漉提取液中有效组分的近红外（NIR）光谱分析方法，可用于中药渗漉提取过程在线分析.以中药三七为对象，采用人参皂苷Rg1、Rb1、Rd的HPLC测定值及三七总皂苷（PNS）的比色法测定值作对照值，建立了NIR光谱与对照值之间的校正模型.比较研究了径向基函数神经网络（RBFNN）和偏最小二乘回归（PLSR）两种建模方法，其中RBFNN校正模型对Rg1、Rb1、Rd和PNS 4种组分的交叉验证均方差（RMSECV）分别为1.120、1.230、0.267、4.749，预测均方差（RMSEP）分别为0.677、0.969、0.155、8.065.研究结果表明，本文方法方便、准确、无损，可推广用于中药提取过程分析及质量控制.

关键词:

近红外光谱, 径向基人工神经网络, 中药, 渗漉提取