National Technical University of Athens (NTUA)
Εντοπισμός μεταβολών σε δορυφορικά δεδομένα πολύ υψηλής ανάλυσης: συνδυάζοντας τεχνικές αντιστοίχισης, κατάτμησης και ανίχνευσης μεταβολών
Abstract
dc:descriptionNowadays, several new satellite and micro satellite constellations are providing plenty of low cost very high resolution data which can observe, with high temporal frequency and spatial resolution, extended regions of the earth. This stream of spaceborne information can significantly contribute to several crucial engineering and environmental monitoring problems though the detection and recognition of multitemporal changes. However, although change detection is a heavily studied problem, it is not a trivial one and most existing approaches address it through step by step procedures with semi-automated or automated registration, segmentation and change detection algorithms.To this end, in this thesis we introduced a single framework which combines energy terms for tackling image registration, semantic segmentation and change detection tasks, through data-driven costs regarding the semantic segmentation likelihoods (various classification approaches can be employed), registration metrics ( e.g., similarity metrics) and change detection scores. In such a joint framework, all tasks are addressed under a fruitful synergy and interaction resulting to a more generic, yet robust formulation, when comparing with the standard step by step procedures and processing pipelines of the literature. We evaluated the developed methodology on a number of datasets including very high resolution satellite image, video and benchmark stereo datasets achieving state-of-the-art quantitative results.Further additional contributions of this PhD thesis include the design, development and evaluation of new methodologies for tackling the main subcomponents of the problem and in particular: (i) new descriptor-based and deformable registration frameworks for the efficient co-registration of multispectral and hyperspecral datasets, (ii) new semantic segmentation approach that integrates learning through deep architectures and (iii) new unsupervised change detection framework by coupling image registration and change detection priors. All developed methods were quantitatively validated against the state-of-the-art showcasing highly promising results.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2017
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Vakalopoulou, Maria
- Βακαλοπούλου, Μαρία
Subjects
dc:subject × 19- Τηλεπισκόπηση
- Τεχνητή νοημοσύνη
- Όραση υπολογιστών
- Παρακολούθηση διαχρονικών μεταβολών
- Οπτικά πολυφασματικά δορυφορικά δεδομένα
- Ανίνευση κτιρίων
- Remote sensing
- Machine learning
- Computer vision
- Monitoring multitemporal changes
- Building detection
- Φυσικές Επιστήμες
- Γεωεπιστήμες και Επιστήμες Περιβάλλοντος
- Επιστήμες Μηχανικού και Τεχνολογία
- Άλλες Επιστήμες Μηχανικού και Τεχνολογίες
- Natural Sciences
- Earth and Related Environmental Sciences
- Engineering and Technology
- Other Engineering and Technologies
Rights
dc:rights- Statement dc:rights
-
- BY_NC_SA
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/41274