The Bayesian network is in fact the Markov blanket of the class node in the classification theme. This

paper deals with a Bayesian network with constraint, in which all feature nodes must be child nodes of class node,

the child nodes can have different relationships with each other, and then some generally-used models of restricted

Bayesian network, namely NB, TAN and BAN, are used to interpret remote sensing images. From the evaluation of the

likelihood function of structures, we can obtain net structure and get the classifier synthesizing the topology and

probability statistical information of feature nodes and the class node. The authors applied these models to the

classification of remote sensing images and discussed their performance.