[visionlist] Call for Participation - CVPR2021 Tutorial Computer Vision and beyond: Pattern Recognition for Satellite Images
Mihai.Datcu at dlr.de
Mihai.Datcu at dlr.de
Mon Jun 14 10:36:37 -04 2021
Computer Vision and beyond: Pattern Recognition for Satellite Images
19th of June
Mihai Datcu, German Aerospace Center (DLR) and University POLITEHNICA of Bucharest (UPB).
Yuliya Tarabalka, LuxCarta Technology, France
Devis Tuia, Swiss Federal Institute of Technology Lausanne (EPFL ENAC)
Satellite or airborne remote sensing are technologies which reached a high level of maturity. Today, the acquired images have extraordinary geometrical and radiometric precision, capturing a wealth of spatial and physical parameters of the observed Earth scenes. The broad open and free access to big remote sensing data volumes creates unprecedented opportunities in applications. Meanwhile, the complexity and volume of the data are asking for sophisticated analysis methodologies. Here, the convergence of the computer vision and remote sensing methods is an important source of new techniques and theoretical perspectives. The tutorial presents an introduction of the remote sensing imagery and mainly the distinct properties in comparison with the ubiquitous pictures. Remote sensing images are frequently acquired in spectral bands as infrared or microwave, thus beyond the human visual perception. Therefore, their analysis needs a translation to the RGB or more advanced visualization methods based on adaptive dimensionality reduction learning projections from multidimensional spaces or interaction in virtual environments. The tutorial addresses these specific topics. An important particularity of remote sensing images is their “instrument” nature, i.e. in addition to the spatial information, they are sensing physical parameters. Here the solutions for information extraction are based on the synergy of machine and deep learning with the physical models of the imaging process, resulting in a new category of hybrid physics aware algorithms. These new paradigms will be introduced and exemplified. The 3D scene reconstruction is one of the important common computer vision and remote sensing objective. The tutorial will introduce the most important methods for precise geometry and temporal analysis of large scale Earth scenes, emphasizing the role of advanced signal processing and AI newest methods. Since the training and benchmark data sets are many times as important as the algorithms them self, their generation and use will be presented. The tutorial concludes with an overview of perspectives and challenges.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the visionlist