The Research On Object-oriented Land Cover Information Extraction

In this paper, the object-oriented software-eCognition is first applied to extract the land cover information of Jiangxi province, and trying to use the user-defined functions as the feature space to assist the land cover information extraction. Finally, the output vector classification result is more easily to combine with ArcGIS. In this experimentation area (Xinyu city, Jiangxi province), four classifiers are applied to extract the information, including Bayes (Bayes classifier). KNN (K-nearest neighbor classifier). SVM (support vector machine), and Decision Tree (decision tree classifier).By comparing the results of these four classification, consider that the object-oriented SVM supervised classification method as more suitable for Jiangxi province land cover information extraction.When extracting the land cover information of2010, at least need use two scene images:the image of spring or summer and the image of autumn or winter. Using the30m resolution DEM for reference, a higher accurate object boundary acquired. When extracting the land cover information in the region covered by the cloud, classify the cloud-covered region as a object of new classification, then try to use other none-cloud images to classify the true land cover of the cloud region. When extracting the wetland information (lakes, ponds, river), the objects are first divided all the water information into pond type, merge this type into large object, then extract the exact river and lake types from the pond types. Finally, the classification result with higher accuracy is achieved, and work efficiency is improved.On the whole, the achievement of this research has a great value of being a reference for our country’s land cover information extraction and the remote sensing information classification.