科研之友
微信
新浪微博
Facebook
分享链接
摘要
The potential of using partial least square based uninformative variable elimination algorithm (UVE(PLS)) on successive projections algorithm (SPA) for spectral multivariable selection was evaluated. A case study was done on the visible and shortwave-near infrared (Vis-SNIR) spectroscopy for the rapid and non-destructive determination of protein content in dried laver. Three calibration algorithms, namely multiple linear regression (MLR), partial least square regression (PLS) and least-square support vector machine (LS-SVM), were used for the model establishment based on the selected variables of SPA, UVE(PLS) and UVE(PLS)-SPA, respectively. A total of 175 samples were prepared for the calibration (n = 117) and prediction (n = 58) sets. The performances of different pretreatments were compared. Both linear calibration algorithms of MLR and PLS and non-linear calibration algorithms of LS-SVM with linear kernel and RBF kernel obtained similar results based on certain variable selection strategies of SPA, UVE(PLS) and UVE(PLS)-SPA. The average improvement percentage of RPD values of four calibration algorithms was 38.66% by calculating SPA on UVE(PLS) processed variables. Therefore there was much improvement of using UVE(PLS) on SPA spectral multivariable selection with both linear and nonlinear calibration algorithms in this case. Moreover, the RPD values of both linear and non-linear models based on the thirteen selected variables of UVE(PLS)-SPA show that coarse quantitative predictions of the protein determination in dried laver is possible based on Vis-SNIR spectra. We hope that the results obtained in this study will help both further chemometric (multivariate selection and calibration analysis) investigations and investigations in the sphere of applied vibrational (Near infrared, Mid-infrared and Raman) spectroscopy of sophisticated multicomponent systems.
