« IEEE fyrirlestur: Ný úthljóðsmyndtækni til greiningar á hjartasjúkdómum | Main | IEEE fyrirlestur: Neural networks for classification and change detection »
IEEE fyrirlestur: Advances in Spectral-Spatial Classification of Hyperspectral Imagery
Skrifað af mou | October 21, 2009
Mánudaginn 26 október klukkan 16:40 í stofu 157 í VR II, Háskóla Íslands mun Yuliya Tarabalka halda fyrirlestur sem nefnist Advances in Spectral-Spatial Classification of Hyperspectral Imagery. Fyrirlesturinn verður fluttur á ensku.
Útdráttur/abstract:
Hyperspectral imaging provides rich spectral information for every
pixel in a particular scene, hence increasing the ability to
distinguish physical structures in the scene. However, a large number
of spectral channels presents challenges for image classification.
While pixel-wise classification techniques process each pixel
independently without considering information about spatial
structures, further improvements can be achieved by the incorporation
of spatial information in a classifier, especially in areas where
structural information is important to distinguish between classes.
In the talk, novel strategies for spectral-spatial classification of
hyperspectral images are introduced. One of the recently proposed
approaches consists of performing image segmentation in order to use
every region from the segmentation map as an adaptive spatial
neighborhood for all the pixels within this region. Furthermore, this
spatial information must be combined with the available spectral
information with a classifier. A simple approach leading to good
classification performances consists of performing pixel-wise
classification and segmentation independently, and then assigning all
the pixels of every region to the majority class within this region
(majority vote rule). A Support Vectors Machine (SVM) classifier has
shown to be very suitable for classification of hyperspectral data.
Therefore, the SVM can be used as a pixel-wise classifier in the
described spectral-spatial classification scheme.
In the talk we also concentrate on approaches to reduce
oversegmentation in an image, which is achieved by introducing
“markers” in the segmentation procedure. We propose to perform a
probabilistic pixel-wise classification first, in order to choose the
most reliable pixels as markers of spatial regions. Furthermore,
marker-controlled region growing is performed, using either watershed
or Minimum Spanning Forest methods. The developed
techniques significantly decrease oversegmentation, improve
classification accuracies and provide classification maps with more
homogeneous regions, when compared to previously proposed
classification methods.
—–
Yuliya Tarabalka is a Phd candidate at Faculty of Electrical and Computer Engineering, University of Iceland, and Gipsa-Lab, Grenoble Institute of Technology, France.
Topics: Óflokkað | No Comments »