Near-infrared spectra of Penicillium camemberti strains separated by extended multiplicative signal correction improved prediction of physical and chemical variations.

Different methods for spectral preprocessing were compared in relation to the ability to distinguish between fungal isolates and growth stages for Penicillium camemberti grown on cheese substrate. The best classification results were obtained by temperatureand wavelength-extended multivariate signal correction (TWEMSC) preprocessing, whereby three patterns of variation in nearinfrared (NIR) log(1/R) spectra of fungal colonies could be separated mathematically: (1) physical light scattering and its wavelength dependency, (2) differences in light absorption of water due to varying sample temperature, etc., and (3) differences in light absorption between different fungal isolates. With this preprocessing, discriminant partial least squares (PLS) regression yielded 100% correct classification of three isolates, both within the cross-validated calibration set and in two independent test sets of samples.