Comparison of Pixel-based and Object-oriented Knowledge-based Classification Methods Using SPOT5 Imagery

Land cover mapping is very important for evaluating natural recourses, understanding the societal and business activities. The remote sensing techniques provide effective and efficient methods to create such maps. To high spatial resolution imagery such as SPOT5 imagery, the land cover classification precision will be improved with the knowledge, Digital Elevation Model (DEM) data and the spatial information such as texture and. Both the pixel-based classification method based on the knowledge rule and the object-oriented fuzzy classification methods have been studied in this paper using SPOT5 high spatial resolution imagery. Some GIS dataset and the texture are integrated into the two knowledge-based classifications in this paper. And the result of accuracy assessment indicates that the two classifications can catch good classification precision, but the objected-oriented classification method does better. Besides, the shape and context information can be used fully to distinguish the roads from the buildings with the objected-oriented fuzzy classification method, which is hard to accomplish in pixel-based classification. Furthermore, the objected-oriented classification method is more suitable in land cover mapping due to its meaningful objects. Key-Words: SPOT5 imagery, Texture, DEM, Pixel-based, Object-oriented, Knowledge-based rules