Skip to Main content Skip to Navigation
Theses

Méthodes de sélection de variables appliquées en spectroscopie proche infrarouge pour l'analyse et la classification de textiles

Abstract : Multivariate analysis methods enable to extract information from spectroscopic data for the prediction of properties of interest. Due to the dimensionality of the data in near infrared spectroscopy, a selection of spectroscopic variables and samples is necessary in order to improve performances, model robustness or to use a simplified instrumentation. Determining the composition of textile is an essential topic due to the wide range of applications. The first study relates to the determination of the cotton content in cotton/polyester and cotton/viscose blend. In order to improve the predictive capacities obtained on the full spectra, two procedures of variable selection, mutual information and the genetic algorithms were applied. The standard error of prediction obtained for the data set cotton/polyester is 2.53% on the 8 variables selected by mutual information. The second part develops the qualitative analysis for the classification of textile samples in three classes according to the physicochemical property of interest. The method of support vector machines has powerful results with a well classified samples rate of 93.2% in prediction. The arbitrary reduction of the number of spectroscopic variables demonstrates that the predictive capacities obtained on the full spectra are not degraded. These results are confirmed by the use of a simplified instrumentation.
Document type :
Theses
Complete list of metadatas

Cited literature [57 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00269380
Contributor : Hervé Vezin <>
Submitted on : Wednesday, April 2, 2008 - 5:30:00 PM
Last modification on : Wednesday, November 25, 2020 - 3:01:19 AM
Long-term archiving on: : Thursday, May 20, 2010 - 9:30:43 PM

Identifiers

  • HAL Id : tel-00269380, version 1

Collections

Citation

Alexandra Durand. Méthodes de sélection de variables appliquées en spectroscopie proche infrarouge pour l'analyse et la classification de textiles. Autre. Université des Sciences et Technologie de Lille - Lille I, 2007. Français. ⟨tel-00269380⟩

Share

Metrics

Record views

937

Files downloads

5429