Skip to Main content Skip to Navigation
Conference papers

Segmentation Sémantique d'Images de Télédétection Combinant Modèles Graphiques Probabilistes Hiérarchiques et Réseaux de Neurones Convolutifs Profonds

Abstract : In this paper, a novel method to tackle semantic segmentation of very high resolution remote sensing data is presented. Deep learning techniques, such as convolutional neural networks (CNNs) and fully convolutional networks (FCNs), have shown exceptional performances in this task. But the accuracy of their classification depends on the quantity and quality of the ground truth used to train them. On the other hand, probabilistic graphical models (PGMs) have sparked even more interest in the past few years, because of the ever-growing availability of very high resolution data and the correspondingly increasing need for structured predictions. The research themes proposed in this paper aim to link and combine different ideas of these approaches (deep learning and stochastic models) to develop new methods of classification of remote sensing images. In order to develop a pipeline combining deep learning and PGM to meet the growing need for precise semantic mapping in remote sensing images, two well-known deep learning architectures such as U-Net and SegNet were considered. The experimental validation was carried out with the “ISPRS 2D Semantic Labeling Challenge” data set on the city of Vaihingen, in some cases with some modifications, in order to approximate the ground truths common in real remote sensing applications, to assess whether the proposed method could improve the accuracy of classification in several cases. The results are significant, because the pipeline studied has a higher recall compared to the standard FCNs considered.
Complete list of metadata

https://hal.archives-ouvertes.fr/hal-03339665
Contributor : Ccsd Sciencesconf.Org Connect in order to contact the contributor
Submitted on : Thursday, September 9, 2021 - 3:41:20 PM
Last modification on : Monday, September 20, 2021 - 2:45:05 PM

File

Pastorino_Martina_ORASIS2021_f...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-03339665, version 1

Citation

Martina Pastorino, Gabriele Moser, Sebastiano B. Serpico, Josiane Zerubia. Segmentation Sémantique d'Images de Télédétection Combinant Modèles Graphiques Probabilistes Hiérarchiques et Réseaux de Neurones Convolutifs Profonds. ORASIS 2021 - 18èmes Journées francophones des jeunes chercheurs en vision par ordinateur, Centre National de la Recherche Scientifique [CNRS], Sep 2021, Saint Ferréol, France. ⟨hal-03339665⟩

Share

Metrics

Record views

57

Files downloads

28