Gram-Schmidt Orthogonalization and Elimination of the Effect of Unwanted Component Spectra Applied to a Biological Mid-infrared Spectra Collection

Abstract

Abstract Near Infrared spectroscopy (NIR) is the most widely used technique for the analysis of major biochemical constituents in food products. The Mid-infrared range spectroscopy is also being more and more studied in the field of food analysis. This range was primarily applied to qualitative analysis of food components, but with the advent of new techniques such as Attenuated Total Reflectance (ATR) together with the possibility of combination with powerful micro-computers, MIR is now more and more used for quantitative analysis. In addition to baseline deformations, interference by unwanted compounds are major sources of problems that are encountered in analyses. We have previously proposed (Cadet et al., 1991) the use of multidimensional statistical analysis combined with Mid-FTIR spectroscopy for the prediction of sucrose concentrations in biological solutions containing three sugars: sucrose, fructose and glucose. In this paper, a least-squares method has been used to assess the elimination of the component spectra associated to the fructose (the unwanted components were first orthogonal zed and normalized by the Gram-Schmidt orthogonalization method). This procedure permitted the automatic subtraction of discriminant spectral patterns representative of fructose concentration before application of Principal Component Analysis (PCA) and Principal Component Regression (PCR). PCA was applied independently before and after spectral correction of the collections. It is found that when the factorial maps obtained before and after correction are compared, the elimination procedure improves significantly the classification of solutions according to their sucrose content. However the bias and standard deviation (SD) values that are associated with the sucrose content predicted values are not influenced by the correction method used: bias and SD are 3.62×10−2 and 3.097×10−1 before correction and after correction they were respectively 3.60×10−2 and 3.104×10−1. This could be explained by the fact that the presence of fructose in solution does not interfere with caracteristic absorption bands of sucrose and that sucrose and fructose concentrations are strongly correlated. The absorbtions bands of the spectral representation of the principal component, which is identified to be that associated with sucrose, are identical before and after correction.