A hierarchical fuzzy clustering algorithm is presented in this paper. As CFCM algorithm provides good classification resolution but the selecting of initial vectors is of great difficulty and blindness. At the first step

an improved SFCM algorithm is adopted for coarse segmentation and its result serves as initial vector for the fine segmentation by CFCM at the second step. Applying this algorithm to snow recognition shows it can yield very satisfying performance in partition resolution while does not reduce computation speed due to a reasonable initial vector.