Skip to Main content Skip to Navigation
Theses

Accélération algorithmique et matérielle des méthodes d’estimation de cartes d’abondances en imagerie hyperspectrale

Abstract : Hyperspectral imaging consists in collecting the reflectance spectrum for each pixel of an image. This measurement technique is used in airborne remote sensing, astrophysics, or microscopy. Processing the large data volume of a hyperspectral image requires a method with both restrained computational cost and limited memory usage. The method proposed in this thesis aims at estimating the abundance maps (component's proportions in each image pixel) by constrained least squares criterion minimization with the addition of a penalization term to ensure the maps spatial regularity. The work done intends to reduce the computing time of an interior point optimization method. Algorithmic modifications based on separable majorization are proposed. It results in a method both faster and more adapted to parallel computing tools. An implementation on Graphics Processing Units (GPU) is achieved and applied in a large scale experiment where a high number of hyperspectral images from Mars Express exploration mission are processed. The developed method is also used in a vegetation monitoring project on the french atlantic coast.
Document type :
Theses
Complete list of metadata

https://hal.archives-ouvertes.fr/tel-01360464
Contributor : Maxime Legendre Connect in order to contact the contributor
Submitted on : Monday, September 5, 2016 - 7:09:27 PM
Last modification on : Wednesday, April 27, 2022 - 3:59:57 AM
Long-term archiving on: : Tuesday, December 6, 2016 - 1:56:01 PM

Identifiers

  • HAL Id : tel-01360464, version 1

Citation

Maxime Legendre. Accélération algorithmique et matérielle des méthodes d’estimation de cartes d’abondances en imagerie hyperspectrale. Traitement du signal et de l'image [eess.SP]. Ecole Centrale de Nantes (ECN), 2015. Français. ⟨tel-01360464⟩

Share

Metrics

Record views

186

Files downloads

210