The increasing availability of high resolution satellite images is an opportunity to detect urban objects such as roads. In order to increasing the precision a new image analysis using object-based approaches has been proposed. In this paper, designing steps of knowledge based of road detection has been presented. In this field, an important challenge is the use of knowledge for automatic road objects identification, and a major issue is the formalization and exploitation of this knowledge. At first, optimum features, including spectral, texture and structural features, are detected using a genetic algorithm with a k-nearest neighbor classifier. After that a rule based road detection strategy has been developed using prior knowledge and optimum features interpretation. The method is designed and validated by IKONOS images of the urban areas of Hobart, Kish and Shiraz. The validation results highlight the capacity of the proposed method to automatically identify road objects using the knowledge based proposed system.