The differences of the spectral curves among different objects are so obvious and important to be extracted through the wavelet transform at different scale. The traditional clustering concepts based on Euclidean distance are also redefined by the wavelet feature correlation coefficient to accurately describe the content of the remotely sensed objects. The fast clustering algorithm for multispectral images based on wavelet feature and simulated annealing increases the number of characteristic points by expanding the spectral bands to enrich the feature of the classes; clustering space is formed by evenly sampled dots; furthermore

simulated annealing leads to the best class centers on the whole scope at a high speed by decreasing the clustering scale and temperature step by step; the class characters is remained strong and durative by choosing the best dot as the class center; it also resolves the initial parameter problem of K means algorithm. The experimental results of Mississippi Thematic Mapper images show that this clustering algorithm is more efficient than other ordinary clustering algorithms such as K means and ISODATA according to the clustering accuracy and speed. Therefore

there are fairly prosperous applications on multispectral images for this fast clustering based on spectral wavelet features extraction and simulated annealing algorithm.