An Automated Procedure for Cluster Analysis of Multivariate Satellite Data
Joakim T. A. Waldemark · International Journal of Neural Systems · 1997
This paper studies the applicability of available cluster analysis methods on multivariate satellite data. Four different methods are studied, these are the Principal Component Analysis (PCA), K-means, Self Organizing Maps (SOM) and the Adaptive Resonance Theory (ART). Special focus is placed on the usefulness of these cluster analysis methods, that is, if the results generated by these methods are relevant and are of help to the physical analysis of the data. A combined SOM adaptive dynamic K-means procedure suitable for automated cluster analysis is presented. This new method is capable of achieving useful partition of multivariate satellite data which has not previously been studied.