Jelenlegi hely

Intézeti szeminárium

Félév: 
2021/22 I. félév
Helyszín: 
Árpád tér 2. Alagsor 6.
Dátum: 
2021-11-16
Időpont: 
14:00-15:00
Előadó: 
Josiane Zerubia (Inria, France, in collaboration with University of Genoa, Italy)
Cím: 
Hierarchical Probabilistic Graphical Models and Neural Networks for Remote Sensing Image Classification: Application to Natural Disasters and Urban Studies
Absztrakt: 

The task of monitoring the Earth’s surface plays an important role in
the framework of the protection from environmental disasters such as
flooding, landslide, or earthquake, and in the field of urban land cover.
Thanks to the substantial amount and variety of information available
from the current space missions, models for multimodal data, typically
multiview, multiscale, and multiresolution methods are becoming more
and more important to face the requirements of remote sensing applications.
In this framework, the challenge is to develop accurate and time-efficient
classification methods, flexible enough to exploit information contained
in multimodal data. In the proposed approaches, multimodal fusion is
addressed by supervised classification methods based on hierarchical
Markov models with a quadtree topology. These models have also been
combined with deep neural networks. The marginal posterior mode (MPM)
criterion is used for inference in the proposed framework. The developed
methods have been experimentally validated with datasets containing
very-high-resolution multispectral, panchromatic, and radar satellite
images, or aerial images. The experimental results suggest that the
methods are able to provide accurate classification maps from input
heterogeneous imagery. The comparison with the state of the art techniques
shows the effectiveness of the proposed approaches.