基于非负矩阵分解的多环芳烃成份识别

大气与环境光学学报 ›› 2015, Vol. 10 ›› Issue (5): 386-391.

基于非负矩阵分解的多环芳烃成份识别

杨瑞芳1,2，赵南京1*，肖雪1，于绍慧3，余晓娅1，刘建国1，刘文清1

(1. 中国科学院安徽光学精密机械研究所中国科学院环境光学与技术重点实验室, 安徽合肥 230031 ;
2.中国科学技术大学，安徽合肥 230022 ;
3.合肥师范学院数学系，安徽合肥　230061)

收稿日期:2014-08-01 修回日期:2014-12-08 出版日期:2015-09-28 发布日期:2015-09-11
通讯作者: 赵南京，男，博士，研究员，博士生导师,主要从事环境污染光学与光谱学监测新技术与方法方面的研究。 E-mail:njzhao@aiofm.ac.cn
作者简介:杨瑞芳(1982-)，女，内蒙古乌兰茬布市人，博士研究生，主要从事光谱学监测方面的研究。
基金资助:
国家自然科学基金(61378041, 61308063)、安徽省杰出青年科学基金(1108085J19)、中国科学院仪器设备功能开发技术创新项目(yg2012071)资助

PAHs Component Recognition Based on Nonnegative Matrix Factorization

YANG Ruifang1,2, ZHAO Nanjing1, XIAO Xue1,YU Shaohui3,YU Xiaoya1, LIU Jianguo1, LIU Wenqing1

(1.key Laboratory of Envirionment Optics and Technology, Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Hefei 230031, China;
2.University of Science and Technology of China, Hefei 230022, China;
3.School of Mathematics and Statistics, Hefei Normal University, Hefei 230061, China)

Received:2014-08-01 Revised:2014-12-08 Published:2015-09-28 Online:2015-09-11

摘要/Abstract

摘要：

从重叠比较严重的混合物三维荧光光谱中恢复单一光谱信号，是光谱解析的难点。考虑到光谱内在的非负性，采用非负矩阵分解的投影梯度和交替最小二乘两种算法，并结合K均值初始化方法，来解析菲、芘、蒽3种芳烃混合物的三维荧光光谱数据，有效避免出现负数的分解结果，提取3种成份的三维荧光光谱,得到计算光谱与对应参考光谱的相似系数均大于0.970。计算结果表明，非负矩阵分解能够克服光谱重叠带来的干扰，有效提取光谱成份，从而实现对菲、芘、蒽的成份识别。其中，交替最小二乘的NMF算法更适合实时在线监测。

关键词: 光谱学, 成份识别, 非负矩阵分解, 三维荧光光谱, 多环芳烃

Abstract:

It is difficult to extract each component from overlapping three-dimensional fluorescence spectra of mixture. Considering intrinsic nonnegativity constraints on spectra, three-dimensional fluorescence spectra data of polycyclic aromatic hydrocarbons (PAHs) mixtures of phenanthrene, pyrene and anthracene is analyzed by using projected gradient and alternating least square algorithms based on nonnegative matrix factorization (NMF) by taking the results of K-means clusting as initial values. The negative data of separated spectra is eradicated. Three-dimensional fluorescence spectra of each component is extracted, and the similarity coefficients between computed spectra and its corresponding standard spectra are computed, which is greater than 0.970. Results demonstrate that three components are recognized accurately by NMF, which could overcome the interference caused by overlapping spectra and extract spectral components effectively. Alternating least square algorithms based on NMF is more suitable for online real-time monitoring.

Key words: spectroscopy, component recognition, nonnegative matrix factorization, three-dimensional fluorescence spectra, polycyclic aromatic hydrocarbons

杨瑞芳赵南京肖雪于绍慧余晓娅刘建国刘文清. 基于非负矩阵分解的多环芳烃成份识别[J]. 大气与环境光学学报, 2015, 10(5): 386-391.

YANG Rui-Fang, ZHAO Nan-Jing, XIAO Xue, YU Shao-Hui, YU Xiao-Ya, LIU Jian-Guo, LIU Wen-Qing. PAHs Component Recognition Based on Nonnegative Matrix Factorization[J]. Journal of Atmospheric and Environmental Optics, 2015, 10(5): 386-391.

参考文献

[1] Stedmon C A, Bro R. Characterizing dissolved organic matter fluorescence with parallel factor analysis:a tutorial [J]. Limnol. Oceanogr.: Methods, 2008, 6: 572-579.
[2] Wang H B, Zhang Y J, Xiao X. Quantification of polycyclic aromatic hydrocarbons in water: a comparative study based on three-dimensional excitation-emission matrix fluorescence [J]. Anal. Sci., 2010, 26(12): 1271-1276.
[3] Wu H L, Shibukawa M, Oguma K. An alternating trilinear decomposition algorithm with application to calibration of HPLC-DAD for simultaneous determination of overlapped chlorinated aromatic hydrocarbons [J]. J. Chemometr., 1998, 12: 1-26.
[4] Lee D D, Seung H S. Learning the parts of objects by non-negative matrix factorization [J]. Nature, 1999, 401: 788-791.
[5] Guillamet D, Vitria J. Introducing a weighted non-negative matrix factorization for image classification [J]. Patt. Recog. Lett., 2003, 24(14): 2447-2454.
[6] Wild S, Curry J, Dougherty A. Improving non－negative matrix factorizations through structured initialization [J]. Patt. Recog., 2004, 37( 11): 2217 -2232.
[7] Li Le, Zhang Yujin. A survey on algorithms of non-negative matrix factorization [J]. Chinese Journal of Quantum Electornics, 2008, 36(4): 737-743(in Chinese).
李乐, 章毓晋. 非负矩阵算法综述 [J]. 量子电子学报, 2008, 36(4): 737-743.
[8] Lin C J. Projected gradient methods for non-negative matrix factorization [J]. Neural Comput., 2007, 19(10): 2756-2799.
[9] Paatero P. A weighted non-negative least squares algorithm for three-way ‘PARAFAC’ factor analysis [J]. Chemometr. Intell. Lab. Syst., 1997, 38(2): 223-242.
[10] Wild S. Seeding Non-Negative Matrix Factorization with the Spherical K-Means Clustering [D]. Colorado: Master’s Thesis of University of Colorado, 2003.
[11] Yu S H, Zhang Y J, Liu W Q, et al. A novel initialization method for nonnegative matrix factorization and its application in component recognition with three-dimensional fluorescence spectra [J]. Molecul. Biomol. Spectr., 2012, 86: 315-319.
[12] Le Qinggui, Liang Guolong. Fast hierarchical alternating nonnegative least squares algorithm for nonnegative matrix factorization [J]. Computer Simulation, 2012, (11): 174-180(in Chinese).
勒庆贵, 梁国龙. 非负矩阵分解的分层最小二乘快速算法研究 [J]. 计算机仿真, 2012, (11): 174-180.