Intra-urban Land Cover Classification from High-resolution Images Using the C4.5 Algorithm

Nowadays, the availability of high-resolution images has increased the number of researches on urban land use and land cover classification. Most of them have used object oriented image analysis with successful results. Although object oriented analysis offers effective tools to represent the knowledge of the scene, the tasks of building semantic network and selecting attributes are time-consuming. These processes demand considerable prior knowledge of the scene and of the urban object characteristics. Therefore, we propose to use the C4.5 decision tree algorithm to help semantic network construction and attribute selection processes. This algorithm selects the best subset of attributes based on an entropy measure and organizes the classes in a decision tree structure. To evaluate the performance of C4.5 algorithm, we conduced a land cover classification in an urban area of São José dos Campos (São Paulo state, Brazil). Two experiments were performed, one based on specialist knowledge using E-Cognition 4.1TM system and the other based on the decision tree generated by C4.5 algorithm. Both provided similar results although the C4.5 experiment was faster than the other. * Corresponding author.