无信息光谱对近红外光谱定量分析的影响  

Effect of uninformative spectra on quantitative analysis of near infrared spectroscopy

作　　者：杜鹏远[1] 郑开逸[2] 杜一平[2] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]华东理工大学上海市功能性材料化学重点实验室,上海200237

出　　处：《计算机与应用化学》2012年第9期1153-1155,共3页Computers and Applied Chemistry

摘　　要：近红外光谱数据因为样品含有相同或高度相似的基体,往往具有相似的光谱形状,而某些光谱可能只是反映了基体信息,而与被测物含量没有相关性,称之为无信息光谱。因为光谱的高相似性,无信息光谱很容易混杂在光谱数据中。论文采用玉米近红外光谱数据,通过向其中人为添加无信息光谱数据,研究无信息光谱对近红外光谱定量分析的影响,采用偏最小二乘法,并结合去一交互检验方法建立分析模型。研究发现,当数据集中含有无信息光谱时,所建立的近红外光谱定量分析模型交互检验误差会明显增大,而且无信息光谱数目越多误差越大。所建立模型对独立检验集的预测误差也明显增高。但简单地从光谱和交互检验结果,或者主成分分析都很难发现和鉴别无信息光谱。As containing same or high similar matrix, near infrared spectra of real samples normally are very similar in shape. Some of the spectra may reflect information about the matrix but not relate to contents of analyte, which are called uninformative spectra. Because of the high similarity between normal and uninformative spectra the uninformative spectra are not easy to be found generally. In the present study, corn NIR spectra are used to investigate effect of the uninformative spectra on models built with partial least squares and leave-one-out cross validation with a method of adding simulated uninformative spectra to the normal spectra. The results show that when the data set contains uninformative spectra root mean squared error of cross validation will increase clearly, and it will also increase with the increase of number of uninformative spectra, while the predicted error for the independent test set is increase significantly. However, it is hard to find and identify the uninformative spectra from the spectral data in view of the spectra, or cross validation plot, and even principal component analysis.

关 键 词：近红外光谱:无信息光谱 偏最小二乘 主成分分析 

分 类 号：O657.3[理学—分析化学]