Skip to Main content Skip to Navigation
Theses

Classification de données massives de télédétection

Nicolas Audebert 1, 2
2 OBELIX - Environment observation with complex imagery
UBS - Université de Bretagne Sud, IRISA-D5 - SIGNAUX ET IMAGES NUMÉRIQUES, ROBOTIQUE
Abstract : Earth Observation allows us to modelize and understand the evolution of our planet. The profusion of aerial and satellite remote sensing images induces the need for automated tools able to semantize such raw data in order to map the Earth. This thesis studies the design, implementation and validation of machine learning strategies, specifically deep convolutional neural networks, for image understanding and automatic mapping. We introduce models for automated interpretation of color, multispectral and hyperspectral images, that are able to exploit spatial relationships between geometrical entities and to produce high precision maps relevant for object detection. We design data fusion architectures using multimodal learning and residual correction that can leverage ancillary data, such as digital surface models and prior geographical knowledge. Finally, we study the generalization abilities of those networks for extreme cases of both limited and very large datasets. All along this work, we thoroughly validate our contributions on various aerial and satellite datasets for land cover and land use classification, building footprints extraction and vehicle detection.
Document type :
Theses
Complete list of metadata

Cited literature [499 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/tel-01960350
Contributor : Véronique Soullier <>
Submitted on : Wednesday, December 19, 2018 - 1:03:34 PM
Last modification on : Tuesday, March 16, 2021 - 3:42:13 PM
Long-term archiving on: : Wednesday, March 20, 2019 - 7:14:00 PM

File

DTIS18247.1541495775.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-01960350, version 1

Collections

Citation

Nicolas Audebert. Classification de données massives de télédétection. Intelligence artificielle [cs.AI]. Université Bretagne Sud, 2018. Français. ⟨tel-01960350⟩

Share

Metrics

Record views

509

Files downloads

4647