光谱预处理在农产品近红外模型优化中的应用研究  

Application of Spectrum Preprocessing Algorithms in Optimizing NIR Models of Agricultural Products

作　　者：吴静珠[1] 李慧[1] 王克栋[1] 陈媛媛[1] 徐云[2] 

机构地区：[1]北京工商大学信息工程学院,北京100037 [2]中国农业大学信息与电气工程学院,北京100083

出　　处：《农机化研究》2011年第3期178-181,共4页Journal of Agricultural Mechanization Research

基　　金：北京市优秀人才资助项目(20081D0500300130)

摘　　要：在农产品品质的近红外光谱分析中,光谱预处理方法的选择直接影响所建品质模型的稳健性和准确性。为此,使用132个小麦样品的近红外光谱集为例,分别对于原始光谱集和添加不同比例高斯白噪声的光谱集建模,比较了几种常见的光谱预处理方法(小波消噪、光程校正、导数、基于遗传算法的谱区挑选、正交信号校正)对建模的影响。从实例中得出:在光谱信号较好的前提下,谱区挑选是首选的预处理方法;矢量归一化预处理能够比较有效地去除高斯白噪声;在噪声较大的情况下,预处理方法组合明显提高了模型的质量,但是组合顺序还有待于进一步研究。Spectral data pretreatment play an important role in near infrared spectroscopy analysis for agricultural products quality detection.This paper compared some representative preprocessing algorithms（wavelet transform,orthogonal signal correction,path length correction,derivative,wavelength selection）by an example.the result indicates that wavelength selection is the first and important step to adapted to the high quality spectrum,normalize algorithm can eliminate gauss white noise effectively,the combination of preprocessing algorithms can evidently improve the NIR model＇s quality while spectrum set has much noise,but the combination order and algorithms selection is needed to study on.

关 键 词：近红外 光谱预处理 模型优化 农产品 

分 类 号：S119[农业科学—农业基础科学]