基于玉米近红外光谱的CWT-SVR蛋白质预测模型  

Study on CWT-SVR Prediction Model for Determination of Protein in Maize Based on Near-Infrared Spectroscopy

作　　者：段凌瑶[1] 侯超逸[1] 李静[1] 赵亚亚[1] 陈士林[1] 侯振雨[1] 

机构地区：[1]河南科技学院化学化工学院,河南新乡453003

出　　处：《平顶山学院学报》2017年第5期50-53,共4页Journal of Pingdingshan University

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个玉米样品的蛋白质含量,并同时进行红外光谱(NIRS)测定.在连续小波变换(CWT)处理玉米NIRS数据的基础上,通过多次随机选择定标集和校正集样品的模型优化方法,建立了二次ε不敏感损失函数的CWT-SVR蛋白质预测模型.模型用于检验集样品中蛋白质的测定,结果满意,且优于传统的偏最小二乘模型(PLS).Protein content 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,the protein prediction model of CWT-SVR of square ε-insensitive loss function were established,based on continuous wavelet transformation process the NIRS data of maize.The results,which were obtained from the measurement of protein content of testing set by using the new prediction model,were satisfactory,better than partial least squares （PLS）.

关 键 词：玉米 蛋白质 近红外光谱 支持向量回归 连续小波变换 

分 类 号：O657.37[理学—分析化学]