moteur de recherche

Séminaire A3SI - Deise Santana Maia
Séminaire A3SI - Deise Santana Maia
15-Apr-2021 13:30
Age: 6 days

Title: Watershed-based attribute profiles for pixel classification of remote sensing data Speaker: Deise Santana Maia

When & where: April 15th, 1:30PM,

Abstract: In this presentation, we introduce the combination of two well-established mathematical morphology notions: watershed segmentation on graphs [1] and morphological attribute profile (AP) [2], a multilevel feature extraction method commonly applied to the analysis of remote sensing images. To convey spatial-spectral features of remote sensing images, APs were initially defined as sequences of filtering operators on the max- and min-trees computed from the original data. Since its appearance, the notion of APs has been extended to other hierarchical representations including tree-of-shapes and partition trees such as alpha-tree and omega-tree. In this context, we have proposed a novel extension of APs to hierarchical watersheds. Furthermore, we extend the proposed approach to consider prior knowledge from training samples similarly to [3,4], leading to a more meaningful hierarchy. More precisely, in the construction of hierarchical watersheds, we combine the original data with the semantic knowledge provided by labeled training pixels. Then, we illustrate the relevance of the proposed method with an application in land cover classification using optical remote sensing images, showing that the new profiles outperform various existing features.

[1] Cousty,  J.,  Bertrand,  G.,  Najman,  L.,  Couprie,  M.:  Watershed  cuts:  Minimum spanning forests and the drop of water principle. IEEE PAMI 31(8), 1362–1374 (2008)
[2] Dalla Mura, M., Benediktsson, J.A., Waske, B., Bruzzone, L.: Morphological at-tribute profiles for the analysis of very high resolution images. IEEE TGRS 48(10), 3747–3762 (2010)
[3] Derivaux, S., Lefèvre, S., Wemmert, C., Korczak, J.: Watershed segmentation of remotely sensed images based on a supervised fuzzy pixel classification. In: IEEE IGARSS. pp. 3712–3715 (2006)
[3] Lefèvre, S., Chapel, L., Merciol, F.: Hyperspectral image classification from multiscale  description  with  constrained  connectivity  and  metric  learning.  In:  2014 WHISPERS. pp. 1–4. IEEE (2014)
Bio: Deise Santana Maia obtained her PhD in Computer Science (2016-2019) at Université Paris-Est, co-advised by Laurent Najman, Benjamin Perret and Jean Cousty. During her PhD, she worked on the theory of hierarchical watersheds on graphs and on their applications to image segmentation. Since november of 2019, she is a postdoc researcher at the OBELIX research team of IRISA, at Université Bretagne Sud. Her current research interests include unsupervised and weakly supervised object detection, deep learning and applications of mathematical morphology to the classification of remote sensing images.

<- Back to: Accueil