IMIS | Flanders Marine Institute
 

Flanders Marine Institute

Platform for marine research

IMIS

Publications | Institutes | Persons | Datasets | Projects | Maps
[ report an error in this record ]basket (0): add | show Printer-friendly version

Assessment of unsupervised classification techniques for intertidal sediments
Ibrahim, E.; Adam, S.; Monbaliu, J. (2009). Assessment of unsupervised classification techniques for intertidal sediments, in: Maktav, D. (Ed.) Remote sensing for a changing Europe: Proceedings of the 28th Symposium of Remote Sensing Laboratories, Istanbul, Turkey, 2–5 June 2008. pp. 348-355
In: Maktav, D. (Ed.) (2009). Remote sensing for a changing Europe: Proceedings of the 28th Symposium of Remote Sensing Laboratories, Istanbul, Turkey, 2–5 June 2008. IOS Press: Amsterdam. ISBN 978-1-58603-986-8. 648 pp., more

Available in Authors 
    VLIZ: Open Repository 157843 [ OMA ]

Keywords
    Clustering; Imaging; Intertidal sedimentation; Spectroscopy; Marine

Authors  Top 

Abstract
    The aim of this study is to explore three techniques for unsupervised classification of airborne hyperspectral imagery of intertidal flats. The unsupervised classification techniques considered are k-means (hard clustering), the Gustafson-Kessel algorithm (Fuzzy clustering), and the mixture of Gaussians model (probabilistic clustering). The behavior and suitability of these techniques is analyzed for sediment classification. Artificial data sets based on real airborne and field spectra are used for this purpose. The sensitivity of the techniques is investigated on two spectral aspects: the effect of within class (intra-class) variability and the effect of spectral dimensionality using feature selection. This sensitivity is expressed as classification accuracy in terms of the Kappa statistic (?) that indicates how better the classification is than chance agreement. The results show that the three techniques are suitable for sediment classification. When there is no feature selection involved, the mixture of Gaussians results in the best classification results. When feature selection is considered, sediment classification accuracy increases for all three techniques applied on the artificial imagery.

All data in IMIS is subject to the VLIZ privacy policy Top | Authors