|本期目录/Table of Contents|

光谱数据分析中的通用非线性变量筛选新方法(PDF)

《纳米技术与精密工程》[ISSN:1672-6030/CN:12-1351/O3]

期数:
2015年3期
页码:
226-231
栏目:
精密测量
出版日期:
2015-05-15

文章信息/Info

Title:
 Development of a Novel Nonlinear Variable Selection Strategy
for Spectral Analysis
作者:
 陈达1 闫孟雨1 李奇峰1 于苓2 金涌3 徐可欣1
 1.天津大学精密仪器与光电子工程学院,天津300072;
2.上海烟草(集团)公司技术中心,上海 200082; 3.中国检验检疫科学研究院,北京 100176
Author(s):
 Chen Da1 Yan Mengyu1 Li Qifeng1 Yu Ling2 Jin Yong3 Xu Kexin1
 1.School of Precision Instrument and OptoElectronics Engineering, Tianjin University, Tianjin 300072, China; 2.Technology Center,

Shanghai Tobacco (Group)Corporation,Shanghai 200082, China; 3.Chinese Academy of Inspection and Quarantine, Beijing 100176, China
关键词:
 非线性变量筛选平均影响值支持向量机集成策略
Keywords:
 nonlinear variable selection mean impact value support vector machines integration strategy
分类号:
R460.4035
DOI:
10.13494/j.npe.20140125
文献标识码:
A
摘要:
 光谱技术在移动互联时代获得飞速发展,但是光谱数据往往存在高维数、多样性等特点,如何对这些复杂多变的光谱数据进行高效分析迫在眉睫.在光谱分析中,变量筛选占有重要

地位,可有效降低光谱维度,并显著提升光谱分析的精度和可靠性.本文发展了一种基于平均影响值支持向量机(mean impact valuesupport vector machines,MIVSVM)的非线性变

量筛选方法,同时兼顾样本分布和非线性因素对变量筛选的影响,有望大幅度提升光谱数据的处理效率.MIV算法有机结合了SVM,采取迭代策略以实现边建模边变量筛选的目的,高效避免了

非线性模型对样本分布的干扰.为验证算法的有效性,将该方法应用于不同数据结构的多组光谱数据定量分析.结果表明,MIV算法有效提升了SVM模型性能,不仅能准确提取重要变量,还能

在保证模型预测精度的前提下提高模型的稳定性,为光谱数据分析提供了一种通用的非线性变量筛选算法.

Abstract:
 Spectroscopy developed rapidly during the mobile internet era, resulting in explosive growth of spectral data. It is thus extremely urgent to process these

spectral data efficiently. In spectral data analysis, variable selection plays a significant role in extracting essential information from the fluctuant complex

spectra to enhance the performance of spectroscopy. In this paper, a novel nonlinear method based on the mean impact value (MIV) and support vector machines (SVM) is

proposed, making a balance between the effect of wavelength and sample space in the presence of nonlinear interference. In the MIV-SVM strategy the iterative

strategy is adopted for simultaneous calibration and variable selection to avoid the interference of nonlinearity. To verify the validity of this method, spectral

data sets with different data structures were adopted. The results indicate that the MIV-SVM strategy is a promising tool for constructing a more robust and

parsimonious regression model compared with routine linear variable selection methods, thus providing a general nonlinear variable selection strategy for spectral

data analysis.

参考文献/References

-

备注/Memo

备注/Memo:
 收稿日期: 2015-04-07.
基金项目: 首都科技条件平台资助项目(Z131110000613066);国家自然科学基金资助项目(21305101, 61378048).
作者简介: 陈达(1979—), 男, 博士,教授.
通讯作者: 陈达, dachen@tju.edu.cn.
更新日期/Last Update: 2015-05-22