Abstract: The classification based on cloud model always uses cloud transform and Pan-Concept-Tree，it brings on inconsistency between the result and prior knowledge.The cloud model with the view of granularity computation was analyzed and the principle of granularity in cloud model was discovered.An approach was proposed for classifying objects in different granularity based on disassembly cloud，and the algorithm description of the approach was given.The algorithm was tested with the data collected by Fengyun-2C satellite，the experimental result showed that the new approach can get quite good effect of classification.