基于NIR、MIR光谱和化学计量学的松花粉、大黄质量控制

【作者】 王亚敏；

【导师】 张卓勇；

【作者基本信息】 首都师范大学 ， 分析化学， 2005， 硕士

【摘要】 近年来,多种现代仪器分析方法被用于中药鉴定和质量控制,红外光谱(IR)法在物质的定性、定量分析和鉴别中一直起着十分重要的作用;作为一种简单、快速、无损的检测手段,近红外光谱(NIRS)技术已成为中药材定性判别和定量分析的新兴方法。近红外光谱区的吸收强度低、谱带复杂、重叠严重,无法使用经典定性、定量方法,必须借助于化学计量学中的多元统计、曲线拟合、聚类分析、多元校准等方法定标,将其所含的定性、定量信息提取出来。人工神经网络(ANN)作为一种智能型算法,具有很强的非线性映射能力,在非线性多元校正中已显露出一定的优势。 大黄是我国著名的特产药材,用药历史悠久,从古至今一直是临床最常用药物之一,以泻下、健胃而著称于世。药典收载的正品大黄为掌叶大黄、唐古特大黄和药用大黄。除正品大黄外,大黄属的其他种在不同地区及民间也做药用,但药效不同,泻下作用差,且有的可引起腹痛,为了确保正品大黄临床用药的安全有效,对大黄药材进行质量控制至关重要。本文第二章将多层感知器神经网络与近红外光谱法相结合对正品和非正品大黄进行了鉴别分析,多层感知器神经网络分别采用了BP算法、DBD算法和QP算法。为了验证神经网络建立的分类模型,本文采用交叉验证方法。预测结果的判定阈值设为0.8和0.2,对三种不同的神经网络算法在大黄鉴别分析中的应用进行了比较研究,其中DBD神经网络的正确识别率最高为94.2%,而BP神经网络和QP神经网络用于大黄分类时正确识别率均为92.3%。结果表明采用以上三种算法的多层感知器神经网络均可用于大黄的分类。 迄今为止的研究结果表明,大黄所含化学成分多达180种以上,化学成分结构明确的有160多种,主要分为蒽衍生物类、萘衍生物类、吡喃酮类衍生物、芪类、苯丁酮类、鞣质类等。药理作用清楚或正在研究中的化学成分有30多种,因此测定大黄中的主要活性成分的含量,作为评价大黄品质的一种科学手段具有十分重要的意义。本文第三章将多层感知器神经网络与近红外光谱法相结合对大黄有效成分蒽醌及其单糖甙类、水溶性蒽甙类、芪甙类和鞣质及其有关化合物的含量进行了快速检测。多层感知器神经网络分别采用了BP算法、DBD算法和更多还原

【Abstract】 In recent years, many modern instrumental analysis methods have been used in identification and qualification of herbal medicine. Infrared (IR) spectroscopy has an important use in qualitative and quantitative analyses of herbal medicine. Near-infrared spectroscopy (NIRS) technique has become a new method of qualitative identification and quantitative analysis of traditional Chinese medicine as a simple, rapid, undestroyed method. Artificial neural networks (ANNs) have recently gained attention as fast and flexible methods to quality control of traditional medicine. ANNs can be trained to recognize patterns by building nonlinear models. The models can be used to generalize their conclusions and to predict patterns not being previously encountered.Rhubarb is a well known traditional medicinal material used in China as a cathartic and a promoter of gastric health. In the Chinese pharmacopoeia there are only three recorded species, namely, R. palmatum, R. tanguticum and R. officinale, and these are referred to as official rhubarbs. Besides the official rhubarbs, radix and rhizome of other species within the genus Rheum are often used in folkloric medicine in various minority regions. However, the purgative activity of some unofficial rhubarbs is much lower than those of the official rhubarbs and some may even cause stomach ache. To ensure the safety and curative effectiveness of rhubarb in clinical practice, it is important to distinguish between official and unofficial rhubarbs. In the second chapter of the present study, combination of multilayer perceptron neural networks (MLPNNs) and NIR spectroscopy was attempted to identify official and unofficial rhubarbs. The MLPNNs were trained with back propagation (BP), delta-bar-delta (DBD), and quick propagation (QP) algorithms, respectively. The MLPNNs trained with the DBD algorithm achieved the highest recognition accuracy 94.2%, and QP algorithm and BP algorithm achieved the same recognition accuracy 92.3%. The purpose of this work is to provide a comparative study of various training algorithms for MLPNNs for identification of official and unofficial rhubarbs.Along with the decrease of wild official rhubarbs and the discovery of other pharmacological activities except for purgative activity, the studies on chemical constituents of rhubarbs have been paid more and more attention. The clinical application of rhubarbs is not only a medicine with purgative activity but with many applications. The chemical constituents of genus Rheum were confirmed being more than 160 species, which were mainly divided into anthracene derivatives, stilbenes, phenylbutanone glucosides, tannins, naphthalene derivatives and chromone derivatives. In the third chapter of the present study, combination of multilayer perceptron neural networks (MLPNNs) and NIR spectroscopy was attempted to develop a method for determining the chemical constituents anthraquinones, anthranones, stilbenes, tannins and related compounds. The MLPNNs were trained with back propagation (BP), delta-bar-delta (DBD), and quick propagation (QP) algorithms, respectively. On the basis of selection of parameters, the best calibration models were developed to correlate the spectra and the values determined by wet chemical analytical method using three different MLPNNs algorithms. Among the three ANN algorithms, the best models were QP neural network when applied them to predict the contents of four main components anthraquinones, anthranones, stilbenes and tannins, performance were 0.55, 0.73, 0.72 and 0.66, the correlation coefficient were 0.8831, 0.8690, 0.7415 and 0.7918, respectively. The aim of this study is to develop a NIRS method for determining chemical constituents in Chinese rhubarbs and to provide a comparative study of various training algorithms for MLPNNs for determining chemical constituents in rhubarbs.Infrared spectroscopy (IR), scanning electron microscope (SEM) and energy-dispersive X-ray analysis (EDX) were used to analyze nutrients in four different kinds of pine pollen powder samples. Based on the differenc更多还原