Résumé
Near infrared spectrometry allows the measurement of the absorbance or reflectance of the infrared region of the electromagnetic spectrum of a sample whose composition one wishes to determine in part. If the principle is common to the various applications, the spectrum produced depends on the specific conditions of measurement and more particularly on the sample (e‧g. granulometry, moisture content), the environment (e‧g. temperature), the spectrometer (e‧g. model, brand, age) but also on the interaction between these 3 types of variables. If it is possible to minimize the influence of the product and the environment (i‧e. by standardizing the sample preparation and measurement conditions), the spectrum remains the product of the sample/spectrometer pair. This dependence of the calibration on the spectrometer on which the measurements have been performed limits the generality of the results and the possibilities of data exchange in the network. It imposes a fastidious work of standardization associated with a loss of precision as soon as the spectrometer park evolves, that the latter gets old or that the predictive model is used on another spectrometer, even from the same manufacturer. In order to increase the generality of the results, it is necessary to compare the evolution of the spectra of the same products on several spectrometers. The specific objective was to create and fill in a standard multi-machine and multi-product database in near infrared spectroscopy. Four repetition of 66 samples belonging to five sample types (e‧g. yam and cassava flour) were acquired on 8 different spectrometers (TANGO, Lab Spec 5000, LabSpec4, MicroNIR1, MicroNIR2, Quality Spec, NIRSystem5000, Camera FX17). A database of 2640 spectra was built. In the next periods, this database will allow to study the interoperability between measurement devices of the network, especially by using artificial intelligence (e‧g. generative adversial network, variational autoencoder).