The overall visual statistical model is put forward based on the structural model. This model not only considers the local features of the objects but also considers the spacial information between objects. The model of local parts which are chosen depending on application, and the model of connection between parts are built separately which are expressed by graph model. We simplify the problem by minimizing the energy function into MAP problem based on the statistical model. In this paper, this model is applied to the typical object recognition of remote sensing image and this model works well not only on single target detection but also on multiple target localization.