基于人工神经网络的淫羊藿红外光谱的研究  

Infrared Spectroscopy of Epimedium Brevicornum Based on Artificial Neural Network

作　　者：张勇[1] 金向军[2] 谢云飞[2] 赵冰[2] 丛茜[1] 

机构地区：[1]吉林大学地面机械仿生技术教育部重点实验室 [2]吉林大学超分子结构与材料教育部重点实验室

出　　处：《光谱学与光谱分析》2008年第6期1251-1254,共4页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(20473029,20573041,50635030;吉林省科技厅重点项目;教育部回国人员启动基金项目资助

摘　　要：对于不同产地和不同栽培条件的药材,其药效的不同是由于其所含化学成分和各成分含量的比例不同所造成的,这种差异将造成红外图谱的差异。但这些差异非常细微,单纯地从谱图去区分其特征是非常困难的。文章利用傅里叶变换红外光谱,测定了42种来自吉林3个不同产地的淫羊藿样品的红外光谱,并对光谱数据进行了相应的预处理。为了提高神经网络的训练速度,在利用人工神经网络建立模型之前,通过小波变换的方法对光谱变量进行了压缩。同时对建立的模型的相关参数进行了详细的讨论。实验表明,建立的模型能够正确地对42个淫羊藿样品进行产地鉴别,同时避免了传统光谱分析对药材的分离和提取,从而为中药质量的科学控制和现代化管理提供了可靠的依据。Regarding raw drugs of the different habitat and the different cultivation condition, its treatment efficacy is different. This is because they contain different chemical composition and different ingredients content proportion, which causes the difference in their infrared spectra. But these differences are extremely slight, and purely differentiating their characteristics from the infrared spectra is extremely difficult. In the present paper, the samples of epimedium brevicornu from different fields of Jilin province were surveyed by Fourier transform infrared （IR） spectra, and the corresponding pretreatment to the spectra data was carried out. Before establishing model through the artificial neural networks, in order to enhance the training speed of the ANN, the spectra variables were compressed through the wavelet transformation, and the parameters of the ANN model were also discussed in detail. The model can distinguish the producing area of the 42 samples of epimedium brevicornum correctly, avoiding the separation and drawing of raw drugs with traditional spectroscopy analysis at the same time, thus offer an effectively and reli- able basis for the quality controls and modernized management of Chinese medicine.

关 键 词：中草药 淫羊藿 红外光谱 人工神经网络 小波变换 

分 类 号：O641[理学—物理化学]