Characteristic Vector Regression Analysis: An Effective Method for Data Reduction of the Spectra Database

Abstract

On the basis of characteristic vector analysis (CVA) and multiple regression analysis (MRA), a method of characteristic vector regression analysis (CVRA) is proposed for the data reduction of the spectra database. The high correlation among the spectra in the spectra database ensures the feasibility of the method. Studies of Gaussian functions and spectra of groundwaters, as well as spectra of soils and plants, with evidence of a high reduction ratio and good retrieval precision, show the effectiveness of the method. Better retrieval precision can be achieved by combining the characteristic vectors with the mean spectrum for the reconstruction of the original spectra. The method can be extended to perform spectral classification and spectral background correction.