基于光谱信息散度的近红外光谱局部偏最小二乘建模方法  

Spectral-information-divergence based local pls modeling algorithm in near infrared spectroscopy

作　　者：鄢悦 张红光[1] 卢建刚[1] 施英姿[2] 陈金水[1] 

机构地区：[1]浙江大学控制科学与工程学院,工业控制技术国家重点实验室,浙江杭州310027 [2]杭州师范大学,浙江杭州311121

出　　处：《计算机与应用化学》2017年第5期351-355,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金项目(61590925,U1609212,U1509211)

摘　　要：常见的近红外光谱分析技术,一般将欧式距离作为相似性判据,但是在很多情况下并不能真实体现样本间的相似性;同时,线性回归模型无法克服校正样本集光谱数据中非线性以及样本差异大而导致的精度降低问题。针对上述问题,本文首次将光谱信息散度引入到局部建模算法中,以未知样本光谱与校正样本光谱间的光谱信息散度作为样本相似性判据,选取一定数量与待测样本最相似的校正样本组成局部校正子集,建立局部偏最小二乘模型。为了验证算法的有效性,将现有的全局建模算法、基于样本光谱间欧式距离的局部建模算法与本文提出的基于光谱信息散度的局部建模算法应用于猪肉近红外光谱标准数据集。实验结果表明:本文新方法的预测均方根误差(RMSEP)分别比现有的两种算法降低了22.8%与48.7%,克服猪肉近红外光谱的非线性和差异性,在近红外光谱定量分析领域具有良好的应用前景。In near infrared spectroscopy （NIR） quantitative analysis, Euclidian distance is always used as the similarity criterion, but it is unable to reflect significant similarity between spectra. Besides, linear regression model cannot overcome the low-precision problem resulted from nonlinearity in corrected spectra, and difference between sample. To address these issues, we propose a novel local modeling algorithm, which introduce Spectral Information Divergence （SID） into local modeling method for the first time. Using SID as the similarity criterion, a certain number of corrected sample, which is similar to testing sample, is selected to construct local PLS model. In order to measure the performance of this algorithm, Global modeling algorithm, local modeling algorithm based on Euclidian distance, and this method are applied to a group of pork standards NIR dataset. Compared the experiment results with the above two algorithms, this new algorithm, improves the root mean square error of prediction （RMSEP） about 22.8% and 48.7%, and overcomes the nonlinearity and differences in NIR datasets of pork, shows a good application prospect in the field of quantitative analysis of near infrared spectroscopy.

关 键 词：光谱信息散度 局部校正方法 近红外光谱 偏最小二乘 

分 类 号：O657.33[理学—分析化学]