基于玉米近红外光谱和离散小波变换的SVR模型稳健性研究  

Research on robustness of support vector regression model base on near infrared spectroscopy of maize and Discrete wavelet transform

作　　者：段凌瑶[1] 陈闯[1] 李静[1] 赵亚亚[1] 陈士林[1] 侯振雨[1] DUAN Lingyao CHEN Chuang LI Jing ZHAO Yaya CHEN Shilin HOU Zhenyu(School of Chemistry and Chemical Engineering,Henan Institute of Science and Technology,Xinxiang 453003,Chin)

机构地区：[1]河南科技学院化学化工学院,河南新乡453003

出　　处：《河南科技学院学报（自然科学版）》2017年第1期43-47,共5页Journal of Henan Institute of Science and Technology(Natural Science Edition)

基　　金：河南省教育厅重点研究项目(13A150282);河南省科技厅攻关项目(122102310278)

摘　　要：采用国家标准方法测定125个玉米样品中的蛋白质、淀粉和脂肪含量,同时测定玉米样品的近红外光谱(NIRS)数据.采用多次、随机选择定标集和校正集样品的方法,对支持向量回归(SVR)模型的参数进行优化,探讨离散小波变换(DWT)对SVR模型的影响.结果表明:DWT可有效去除玉米NIRS数据中的背景和噪声,建立的DWT-SVR多变量回归模型具有较好的稳健性,可实现玉米样品中蛋白质、淀粉和脂肪的同时测定.The content of protein,starch and fat of 125 maize samples were measured by using the national standard method and near-infrared spectroscopy（NIRS）,simultaneously.By repeated and random selecting the optimization method of model of standard and calibration set,parameters of support vector regression（SVR） model were optimized,the influence of discrete wavelet transform（DWT） on SVR model were also discussed.The results showed that DWT can remove the background and noise in the maize NIRS data effectively,DWT-SVR multivariate regression model had good robustness which can measure the content of protein,starch and fat in maize at the same time.

关 键 词：玉米 蛋白质 近红外光谱 支持向量回归 离散小波变换 

分 类 号：S513[农业科学—作物学]