Transfer of spectral data between Fourier-transform infrared spectrometers for use in discriminant analysis of fruit purees

The mid-infrared spectra of raspberry and strawberry purees were acquired on two Fourier-transform spectrometers. A pre-treatment procedure, involving baseline correction and area normalisation, was developed that made the spectra from the two spectrometers compatible for use in discriminant analyses. The discriminant analyses were performed by subjecting the spectra to a principal component (PC) analysis, grouping them according to fruit species and then assigning samples to the class with the closest group mean in PC space. Models using the first PC scores of the pre-treated spectra from one spectrometer were able to correctly assign the fruit species of between 91.1 and 96.4% of the pre-treated spectra of independent test sets of the other. However, the projection of the spectra from each spectrometer into the same PC space showed differences in the mean positions of equivalent groups along the second and higher PCs. Therefore, the transfer of this data set is only wholly reliable using the first PC scores. Tests with the uncorrected spectra demonstrated their incompatibility through the complete lack of overlap of equivalent fruit species groups in PC space.