Abstract: Traditional unmixing methods are based on the fixed endmember,and need to assume that the remonte sense image has pure pixel.In fact,this assumption is not necessarily true,and all pixels are not composed of the same endmembers．This paper merged the endmember extration and unmixing into one step,and abstracted it to a random process based on a standard spectral library．Within the premise of variable number of endmembers,Reversible jump MCMC method was used to estimate parameters．The accumulated knowledge of endmembers was used during the state transition process,to improve algorithm efficiency．This algorithm does not require human intervention．It can achieve automated unmixing,and has a high accuracy．The experiments show that the algorithm based on MCMC is superior to the traditional unmixing method in both accuracy and stability.

Key words: Unmixing,Spectral library,Variable endmember,Random variables,MCMC