移动窗口偏最小二乘法优选猪油丙二醛近红外光谱波段

Region Optimization in FT-NIR Spectroscopy for Determination of MDA in Lard with Moving Window Partial Least Squares

移动窗口偏最小二乘法（moving window PLS,MWPLS）是1种波长区间优选方法,主要用于选择与待测组分相关的高信息量的光谱区间。将该方法应用于猪油中丙二醛（malondialdehyde,MDA）含量的近红外定量分析模型的构建中,优选波长区间。通过改变MWPLS的窗口宽度,原始光谱以及经S-G卷积求导（一阶和二阶）预处理的近红外光谱,优选出多个与猪油中丙二醛含量相关的区域。评价及验证结果显示,原始光谱不经校正处理,移动窗口宽度为160个波长点时,优选到的光谱区域（7 305.4 cm^-1-6 078.8 cm^-1）所构建的定量分析模型最佳,其主因子数（Rank）、决定系数（R^2）、交互验证的校正标准偏差（RMSECV）、预测标准偏差[RMSEP（mg/100g）]、Bias和交互检验相对标准偏差（RPD）分别为5,0.9944,0.0596%,0.0545,-0.0181和14.4。这说明用MWPLS可以筛选猪油中丙二醛的近红外光谱信息区间,提高定量分析模型的预测能力,并降低数据的处理量（数据点由1 154个减少为162个）。

Moving window partial least squares（MWPLS） is one method of region optimization, which is mostly used in selecting spectral region that including large information related to components to be determined in the samples.MWPLS was used to select appropriate frequency range for set up a quantitative analysis PLS model of malondialdehyde（MDA） in lard in this paper. The Fourier transform near infrared reflectance（FT-NIR） spectra were processed by MWPLS after Savitzky-Golay Derivative（1^stand 2^ndderivative） or not, different spectral regions which is relevant with MDA content in lard were selected by changing window size of MWPLS. The evaluation and validation results show that the best region for setting up a best PLS model is the original spectrum between 7 305.4 cm^-1and 6 078.8 cm^-1, and the corresponding PLS model＇s rank, determination coefficient（R^2）, standard error of cross validation（RMSECV）, standard error of prediction（RMSEP）, bias and residual prediction deviation（RPD） are 5, 0.9944, 0.0596, 0.0545,-0.0181 and 14.4, separately. This suggests that MWPLS is a valid method to look for spectral feature that enhance the quality and reduce spectra data（from 1 154 reduce to 162） of the quantitative analysis model of MDA in lard.