非负矩阵分解在中药鱼腥草近红外光谱定性判别中的应用  

Application of non-negative matrix factorization in discrimination of Houttuynia cordata Thunb.by near infrared spectroscopy

作　　者：曲斌[1] 孟庆华[2] 张婷婷[1] 辛妮[1] 胡育筑[1] 

机构地区：[1]中国药科大学分析化学教研室,江苏南京210009 [2]徐州师范大学化学化工学院,江苏徐州221116

出　　处：《计算机与应用化学》2009年第10期1353-1357,共5页Computers and Applied Chemistry

摘　　要：非负矩阵分解(NMF)是1种全新的提取特征和数据降维方法,本文将NMF率先引入近红外(NIR)光谱的数据处理,以识别中药鱼腥草为例,将收集到的干燥鱼腥草样品粉碎后,取适量置于石英样品杯中,以积分球漫反射法采集样品的NIR光谱。将各样本NIR光谱数据经多元散射校正后,计算和研究鱼腥草NIR漫反射光谱的NMF,实现快速鉴别鱼腥草的质量,并说明NMF分解结果更加体现了基于局部表示的性质,更能代表数据的局部特征。在与传统的统计方法--主成分分析(PCA)比较的过程中,讨论了NMF的多解性问题,提出基于奇异值分解的简洁初始化算法,讨论了NMF低维特征数r的选择问题,并通过PCA选择NMF的低维特征数r值,讨论了矩阵中样本的排列顺序对结果的影响,通过计算说明矩阵中样本的排列次序对识别没有影响。在探讨NMF的优势与不足的原因的基础上,证明NMF是较理想的数据分析方法。Non-negative matrix factorization （NMF） is a novel approach for feature extraction and dimensionality reduction. In this paper, NMF was firstly introduced and proposed for processing NIR spectra. As an example, powered Yuxingcao （ Houttuynia cordata Thunb. ） were collected from different origins and batches. The samples were placed in the quartz sample cup, and NIR spectra were recorded using the integrating sphere module by diffuse reflection, respectively. After preprocessing all NIR spectra by means of multiple scattering correction （ MSC）, the scores plot based on NMF algorithm was developed to discriminate the varieties of these herbs. It is shown that NMF is able to produce a parts-based representation, and can get parts - based features of objects. Detailed comparisons were made between the NMF method and the traditional used principal component analysis （PCA） method. The problem of multiple solutions on NMF was discussed, and singular value decomposition-based simplified initialization algorithm was proposed. The selection of low rank dimensionality r of NMF was discussed, and here PCA was used to estimate r. The influence on the different sequences of samples in the data matrix was also discussed. After calculation, it was indicated that the sequence has no influence on pattern recognition. On the basis of discussion and analysis of the advantages and disadvantages of NMF, it was shown that NMF is a more effective and promising method for rapid identification of the quality of the traditional Chinese medicine （TCM）.

关 键 词：非负矩阵分解 近红外 鱼腥草 识别 

分 类 号：O6[理学—化学]