Séminaire de Norman Kerle (Université de Twente, NL)


 AHLeGall    17/04/2014 : 21:32

norman_kerle.gif

Le jeudi 17 avril 2014 à 14h, en salle de conf de l'OSUR (campus de Beaulieu, bâtiment 14B, RDC), Norman Kerle propose un séminaire intitulé

Le jeudi 17 avril 2014 à 14h, en salle de conf de l'OSUR (campus de Beaulieu, bâtiment 14B, RDC), Norman Kerle propose un séminaire intitulé

Object-based image analysis to study socio-environmental systems

Présentation :
OSUR carries out advanced research on land, ecosystems and society, which implies substantial challenges related to data source selection and information extraction, monitoring, change detection, and dealing with ever more complex systems and inter-system connections. Remote sensing has been shown to be a useful ally in all of the above domains, yet growing complexity in questions addressed have weakened many traditional analysis approaches, in particular those based on pixels. Object-oriented image analysis (OOA) has been gaining prominence due to its ability to mimic human cognitive processing, and its ability to make effective use of process and feature knowledge. With the roots of image segmentation going back decades, only recently developed tools have been allowing an effective knowledge-driven processing of the resulting segments/objects, in any kind of 2D raster imagery or 3D data cubes. This talk provides an overview of the OOA work carried out at ITC. This includes research on relatively simple feature inventorisation (e.g., urban trees), quantification of natural processes (including both state and change, e.g., of landslides and erosion), as well as elements related to human-environmental interaction such as risk and disasters (including hazard indicators, vulnerability, deprivation, or structural damage). Data used have been varied, from both space- and airborne platforms (most recently from unmanned aerial vehicles), and including vertical, oblique and transverse types. The work has also exposed a number of remaining challenges, such as flexible adaptation of developed procedures across different datasets with variable characteristics. Approaches from the machine learning and computer vision communities have helped to overcome some of the problems.


Norman Kerle est Professeur à l’Université de Twente (Faculty of Geo-Information Science and Earth Observation, Enschede, The Netherlands)


Contact OSUR : Thomas Corpetti (LETG-Rennes-COSTEL