One of the unsolved problems of Geographic Object-Based Image Analysis (GEOBIA) is "the classification results may be inconsistent by different expert in the process of image analysis". Based on geo-ontology theory, this paper presents a novel framework "geo-graphical entity description-model building-object classification" to improve the interpretation of GEOBIA results. A geographical entity is expressed formally from the perspective of geo-ontology based on the characteristics of remote sensing image and expert knowledge. The semantic network model is built by using knowledge engineering methods and computer-actionable formal ontology languages. The image objects are classified based on semantic network model and expert rule. In the case of Land-cover classification, results show that, this method can not only obtain the classification results which reflect the geographical objects, but also grasp the semantic information of the geographical entities, and realize the sharing of land-cover classification knowledge and the reusing of the semantic network model. This new approach provides a holistic framework and method for GEOBIA.