The Object-oriented Classification Method In Karst Area Of ​​image Classification

With developed of remote sensing technology, image resolution is higher than before, high-resolution images are used more and more extensive in Land using, exploration, environmental monitoring and so on. Compared with ordinary remote sensing images, high-resolution images contain richer spatial Information. Traditional classification methods, such as supervised classification and unsupervised classification, can are not satisfied the requirements of classifying high-resolution images,and the result is not accurate. In the experiment, eCognition that due to object-oriented classification was utilized to classify high-resolution image and was contrased with the result that using maximum likelihood method. At last Superiorities of object-oriented classification ware proved.The method of object-oriented classification is used in some fields but less in karst areas. Test area is typical karst area Southwest China, underground river of ZaiDi,GuiLin. First, Object-oriented classification has great different with traditional classification method. That is segmentation has been entered into classification Steps. Image segmentation is a step that image is divided into polygons that have different size and disjoint. These polygons have the same spectral characteristics and properties. In the step, different segmentation parameters can be choised in orther to get not the same effect. Second, eCognition is provided a function that establishing structures of class level. These structures are belong to upper and lower class relationship. Characteristics and properties of upper class were inherit by lower classes. On the basis of these features, some classification could be divided more detail and will not interference each other. That increase classifying efficiency. After automatic classifying, the result can be manual modified so as to more coincide with objective facts.In the experiment, maximum likelihood, a universality and representative method, has been used to classify the same image. At last, The two results that utilized different methods were contrasted. It shows traditional method has many limitations in classifying high-resolution images. Wrong-classification and phenomenon of salt and pepper are very serious. The result of Object-oriented classification has high precision, good continuity and clear boundary. The accuracy assessment shows that the new is higher than maximum likelihood's.The result demonstrates that roads, residents, arable lands and vegetatione which grow on different lithologies can be accurate classed by eCognition. Contrasted with the method of maximum likelihood, object-oriented classification has great advantages in classifying high-resolution karst area images. With improving of theories and algorithms, the method will have great potential and good future prospects.