Remote-sensing image characterizes terrestrial object by its spectrum and space identity. The CTBT satellite nuclear test verification and a crop assessment could be carried out by use of multispectral image of particular area in different period. Based on K-L transformation, a clustering algorithm to deal with redundant data between several bands of multispectral images was proposed. A multispectral image of certain nuclear test site and a civilian TM image were processed and analysed. The results indicate that the algorithm is available.