Using Tasseled Cap Transformation and Finite Gaussian Mixture Model to Classify Landsat TM Imagery Data

An unsupervised classification method combining tasseled cap transformation (TCT) and finite Gaussian mixture model (FGMM) for Landsat TM (thematic mapper) imagery data is proposed in this paper. The spectral dimensionality of the imagery data is firstly reduced by TCT into the brightness component (TCTB) and greenness component (TCTG) and wetness component (TCTW), then the transformed data is modeled by FGMM, the parameters of the model are estimated using the expectation-maximization (EM) algorithm. Finally the data after TCT is classified according to the mixture model. The results from the present study suggest that the TCTB is enough to classify the Landsat TM image to water, vegetation and town or bare land, and the combination of TCTB and TCTG is better to classify the image to water, wetland, shrub and grass land, farmland and town or bare land than the combinations of TCTG and TCTW, TCTB and TCTW, and the combinations of TCTB, TCTG and TCTW is the most reasonable and delicate method for the classification of Landsat TM imagery data.