FT-NIR光谱应用于柚子皮果胶定量分析的波段优选

基于傅里叶变换近红外(FT-NIR)漫反射光谱分析技术,采用分段多元散射校正(PMSC)预处理与移动窗口偏最小二乘(MWPLS)方法为柚子皮果胶的定量分析优选波段。基于全谱进行PMSC预处理,在此基础上结合留一交叉检验技术优选21个分段窗口大小,以此设定为MWPLS方法波段优选的窗口宽度。详细讨论先后采用或不采用MSC预处理的MWPLS模型的预测效果,并从中优选定标预测模型。最优的方案为先MWPLS选择波段,后MSC预处理再建立PLS模型,其优选波段为8 540—7 944(cm-1),窗口宽度为310,对应的最优PLS因子数为11,相应的预测均方根偏差(RMSEP)和预测相关系数(RP)分别为0.520 7(%)、0.871 1。该模型明显优于先对全谱做MSC预处理再建立MWPLS模型,且大幅度优于基于原光谱数据直接建立的MWPLS模型。结果表明,采用PMSC预处理为MWPLS方法搜索窗口宽度预设值,大大降低了MWPLS模型的运算量,有利于波段优选光谱。MWPLS模型结合MSC预处理方法可以有效地应用于柚子皮果胶成分的FT-NIR定量分析。

The Waveband Selection for FT-NIR Spectroscopy Applied to the Quantitative Analysis of Shaddock Peel Pectin

Piece-wise multiplicative scatter correction (PMSC) pretreatment and moving window partial least squares (MWPLS) method were applied to the waveband selection for the quantitative analysis of the pectin in the shaddock peel samples based on the Fourier transform near-infrared (FT-NIR) and the diffuse reflectance technology. PMSC pretreatment was employed for data correction on the whole spectral region, and 21 favorable piece windows were selected on the basis of the leave-one-out cross validation technology. Then the waveband width of MWPLS was set changed as these 21 different values. Moreover, the MWPLS models were in detail optimized by comparing the prediction results in the discussion of the three different cases whether or not using the MSC pretreatment and the optimal models were selected. The best case was the PLS model established on the waveband firstly selected out and then pretreated by MSC. The selected waveband was 8540-7944 (cm-1) with the width of 310, the optimal number of PLS components was 11, and the corresponding root mean square error of prediction (RMSEP) and correlation coefficient of prediction (RP) were 0.5207 (%) and 0.8711 respectively. This model is obviously better than that with MSC pretreatment before waveband selection, and far better than that without MSC. Results showed that the PMSC pretreatment is good at searching favorable values for the change of waveband width of the MWPLS, which lowers the computing workload of MWPLS, and profits the waveband optimization. MWPLS modeling and MSC pretreatment can be well applied for the FT-NIR quantitative analysis of the pectin in the shaddock peel.