Vol. 5, No. 2, 146-160, 2006 |
|||||||
Sediment type unsupervised classification of the Molenplaat, Westerschelde estuary, The Netherlands Stefanie Adam, Ilse Vitse, Chris Johannsen and Jaak Monbaliu
Abstract A hyperspectral image of the Molenplaat, an intertidal flat in the Westerschelde estuary, the Netherlands, was acquired with the HyMap sensor in June 2004. The goal of this research is to perform, analyse and evaluate unsupervised classification methods for sediment types on the imagery. The unsupervised methods are based on Principal Component Analysis (PCA) or Iterative Self-Organizing Data Analysis Technique (ISODATA), and consist of three steps: (a) classification into spectrally distinct clusters, (b) post-clustering treatment, and (c) assignment of labels to the clusters. The result consists of 13 clusters after the post-clustering treatment, and of 8 or 9 classes after labelling for either the PCA or ISODATA method. A supervised Spectral Angle Mapper (SAM) classification was performed using field data to evaluate the unsupervised classification results. The labelling of the unsupervised clusters was also partly based on the SAM results, due to limited field data. The comparison of the results reveals that 69% and 73% of the pixels of PCA and ISODATA classification respectively were identically labelled in the supervised classification. Moreover, the mismatches were mainly found in two classes, while the other classes showed high similarities, indicating the plausibility of using unsupervised classification methods for intertidal sediment types. Additional strengths of the unsupervised classification methods are (a) the distinction of classes that were not visited during field work and not classified in the supervised classification, (b) the identification of spectrally distinct areas that should be characterised during field campaigns, and (c) the user-friendliness thanks to limited required field knowledge and short calculation time. | |||||||
View Full Text (pdf file, 2.5 MB) |
![]() |
||||||
History Submitted: 16 Jun 2005 Revised: 06 Mar 2006 Accepted: 18 Apr 2006 Published: 18 May 2006
Citation |
|||||||
|
|||||||
ISSN 1729-3782 |