LARS Tech Report Number
Nonsupervised classification by clustering has been shown to be a very important tool in the analysis of satellite remote sensing data. However, clustering algorithms which use Euclidean distance as a measure of similarity are highly sensitive to scaling differences among the variables which participate in the clustering process. Since the Landsat MSS spectral bands have different ranges and different calibration functions, this scaling sensitivity is likely to have a significant impact on the results of clustering Landsat MSS data, as is demonstrated by the experiments described in this paper. A rescaling strategy for Landsat MSS data is recommended which seems to give appropriate relative weights to the four spectral bands.
Date of this Version