Skip to Main content Skip to Navigation

Représentation sémantique de données géospaciales au service de l'analyse de changements

Abstract : Change detection from Earth observation satellite images is a useful task for monitoring natural or human-induced changes, but also the impact of specific events (fires, floods, etc.). The use of machine learning to detect areas of change is producing increasingly accurate results, without however providing information on the nature or the cause of these changes. In order to meet this type of need, our thesis aims at developing solutions based on formal vocabularies and ontologies, on knowledge representation, annotation and semantic integration of metadata associated with satellite images and more generally with geo-localized data. Indeed, semantization techniques are a way to offer an intelligent interpretation of data. The case study selected to illustrate the contribution of this approach is the monitoring of changes at different time steps and different scales of restitution. The objective is to enrich the metadata from the image streams by associating them with conceptual categories that give them meaning, to couple them with pre-processing that meets the specific requirements of the study of changes and to integrate them with other available geographic information (climatic and meteorological data, demographic data, etc. as needed). The thesis aims to show the contribution of semantic metadata for change monitoring but also to evaluate the scaling of semantization techniques, including data representation in semantic stores.We propose a process that manages the complete cycle of generation and exploitation of knowledge graphs from rasters acquired from remote sensing and data from open data. The innovative features of this process are the following:i. An algorithm for the automatic identification of regions of interest (ROI) associated with similar values of an indicator computed from a satellite image, thus ensuring a precise geographical division as a reference for data integration.ii. A semantic oriented approach for the generation of knowledge graphs from different sources (raster, open linked data and social networks).We validated the approach with several case studies on fire detection that show the added value of these knowledge graphs to identify regions where a strong change has been detected.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Thursday, March 24, 2022 - 11:45:28 AM
Last modification on : Monday, July 4, 2022 - 9:16:39 AM
Long-term archiving on: : Saturday, June 25, 2022 - 6:58:55 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03618363, version 1


Jordane Dorne. Représentation sémantique de données géospaciales au service de l'analyse de changements. Traitement des images [eess.IV]. Université Toulouse le Mirail - Toulouse II, 2021. Français. ⟨NNT : 2021TOU20068⟩. ⟨tel-03618363⟩



Record views


Files downloads