This paper studies the application of multi-channel fusion in the SAR image classification.In order to make full use of the membership information of pixel to the object classes

we use Bayesian fusion to fuse the information on measurement level.As prior probability of every object class is used in fusion

two methods are used to estimate the probability.The estimation causes a kind of mosaic effect

so modulated Gaussian distribution is introduced to deal with the problem but setting higher threshold to end the iteration proves to be a better method.In order to get an overall prior probability for every object class

three fusion methods of prior probabilities are proposed.Finally

a relative optimal method of multi-channel SAR image classification is achieved with experiments and studies.