| In recent years,with the rapid development of China's economy and extensive urban growth,the construction and management of cities are facing more and more severe test.In order to grasp the dynamic changes in the city,we need to know the information of structures timely and accurately.The key technology of urban building extraction is studied with GF-1 satellite remote sensing image as data source.In this study,we adopted the method of object oriented classification which based on knowledge rules.Firstly,the thesis focuses on the research of image segmentation.Against the situation that the over-segment problem of traditional watershed algorithm is very serious and the efficiency of Stoer–Wagner algorithm is badly low,we had studied and improved the method of image segmentation.The watershed algorithm was improved and this method was combined with the Normalized Cut algorithm.Then,the phenomenon of over-segment was eliminated and the efficiency of the algorithm was elevated.Secondly,the thesis studied the method of image classification.This study used the method of the Support Vector Machine to finish the classification.We provided sample data and analyzed the features of those sample data.Then we assessed the importance of these features and judged the situation whether each feature can join the classification of images.When the features had been chosen,it was used to train the classifier.Then the classification was executed with the trained classifier.Finally,the thesis studied the Morphological shape optimization method.The study conducted the hole filling and small area removal of the result of the classification with the Morphological shape optimization method.And we extracted the boundary of buildings by edge detection.After that,we assessed the accuracy of the result.It had been shown that,the method of object oriented classification which based on knowledge rules could get a commendable result in the classification and this study was pretty potential. |
PDF Full Text Request