Abstract
Infrared reflectance spectroscopy is often used to predict the composition of various food products. However the precision of the analysis could be improved. For this purpose, several treatments can be used; (i) baseline correction. Baseline deformation is one of the major problem that is associated with infrared spectra; (ii) spectra can be expanded in terms of a set of orthonormal polynomials derived from the Legendre polynomials, and the leading terms of the expansion, which contains most of the baseline variation, can be removed; (iii) the classification of samples prior to analysis is one way to obtain more homogenuous families. Hence, with the aim of improving the precision of the values obtained during quantitative determination of constituants from biological samples, we have sucessively applied different mathematical treatments on their infrared spectra: baseline correction, Legendre polynomials decomposition and classification procedure. The application of these mathematical treatments improves successively the precision of the predicted sucrose content values in biological samples. The mean and standard deviation of the differences between predicted and reference values are: 1.52×lO−2 and 3.17×10−1 respectively before correction and (i) after the baseline correction: 5.06×10−2 and. 3.08×0−1 % respectively, (ii) after decomposition Legendre polynomilas: -6.13×10−3and 2.76×10−1 % respectively, (iii) and after classification procedure according to correlation distance: -1.00×10−3 and 2.65×10−1 % respectively. Surprisingly, when only classification procedure is used for correction, the mean and SD values are 7.43×10−2 and 2.62×10−1. Hence, it appears that the successive mathematical treatments can be advantageously substituted by a single classification procedure.