Spectrométrie moyen infrarouge dans le lait : contribution à l’étude des facteurs principaux visant à accroître la qualité des modèles de prédiction et à assurer leur contrôle
- Dehareng, F. (2023). Spectrométrie moyen infrarouge dans le lait : contribution à l’étude des facteurs principaux visant à accroître la qualité des modèles de prédiction et à assurer leur contrôle. Gembloux Agro-Bio Tech, University of Liège, 197.
| Type | Thesis |
| Year | 2023 |
| Title | Spectrométrie moyen infrarouge dans le lait : contribution à l’étude des facteurs principaux visant à accroître la qualité des modèles de prédiction et à assurer leur contrôle |
| University | Gembloux Agro-Bio Tech, University of Liège |
| Label | U11-0006-Dehareng-2023 |
| Number of pages | 197 |
| Abstract | The use of Fourier Transform Medium Infrared Spectrometry (FT-MIR) in milk has recently allowed the development of prediction models for a wide range of new phenotypes. To ensure their robustness, it is important to cover as much phenotypic variability as possible. Moreover, the quality of predictions must be ensured by controlling all the factors that can influence it. The objective of this thesis is to synthesize recommendations and practices to ensure the robustness while maintaining the quality of FT-MIR predictions of milk samples analyzed in routine. First, sample selection is important. Two sampling strategies were investigated to increase the variability of the calibration sample set. The first was based on an original method of sorting the reference samples to be analyzed on the basis of their spectral variability (X). A second one aimed at creating variability in the calibration set directly from the phenotype (y). Then, the choice of the reference analysis is an important factor that should not be exclusively based on precision. Indeed, other criteria will have to guide this choice in order to deal with technical, practical, financial and ethical constraints. The use of pre-processing or the choice of algorithm used to build a model will also influence its performance. On the other hand, most of the new models use a larger number of wave numbers, or even the use of indirect correlations. The control of interferences and spectral deviations of the devices will be an important point to master, because any deviation could lead to unfortunate consequences. Thus, the use of adequate spectral standardization methods seems a promising solution. The procedures currently used in routine to validate methods, to ensure their adaptability to a given geographical area, or to control the quality of routine equipment, will have to be adapted or new procedures will have to be created. In conclusion, this thesis work, reflecting 20 years of experience, was able to put forward many solutions and future developments to improve the robustness of the equations and maintain the quality of predictions in routine. The results obtained underline the need to create international collaborations both for the development of predictive models and for issuing recommendations in order to fully use the potentialities of FT-MIR spectrometry for milk analysis. |
| Fichier | |
| Authors | Dehareng, F. |
Team
