An approach for extracting the relevant features for multivariate calibration of the hydroxyl number in a polyol from a nearinfrared (NIR) spectroscopic data set by using the Fourier transform, is presented. It is carried out in the frequency domain starting from the frst 50 power-spectra (PS) coeffcients as the input to a genetic algorithm (GA). The appropriate PS coeffcients selected by the GA were used to build a multiple linear-regression (MLR) model. The performance of the new approach is compared with MLR after wavelength selection with GA, with the standard PCR and PLS methods applied to the wavelength domain, and PCR and PLS applied to the full PS domain. Furthermore, it was also compared to the `Uninformative Variable Elimination' (UVE) PLS method in the frequency domain. The results demonstrate that PS is a fast and powerful reduction method. The coeffcients selected are of two types: one that correlates with the characteristic investigated, and the other that takes into account different clusters. This also shows that the method can be used to investigate the structure of the data.
Application of Fourier transform to multivariate calibration of near-infrared data
PASTI, Luisa;
1998
Abstract
An approach for extracting the relevant features for multivariate calibration of the hydroxyl number in a polyol from a nearinfrared (NIR) spectroscopic data set by using the Fourier transform, is presented. It is carried out in the frequency domain starting from the frst 50 power-spectra (PS) coeffcients as the input to a genetic algorithm (GA). The appropriate PS coeffcients selected by the GA were used to build a multiple linear-regression (MLR) model. The performance of the new approach is compared with MLR after wavelength selection with GA, with the standard PCR and PLS methods applied to the wavelength domain, and PCR and PLS applied to the full PS domain. Furthermore, it was also compared to the `Uninformative Variable Elimination' (UVE) PLS method in the frequency domain. The results demonstrate that PS is a fast and powerful reduction method. The coeffcients selected are of two types: one that correlates with the characteristic investigated, and the other that takes into account different clusters. This also shows that the method can be used to investigate the structure of the data.I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
