标题：High-Resolution Satellite Imagery Analysis Based on Objected-Oriented Technique- the Control Zone Exaction in "Four Mountain" Region of Chongqing for Example
期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2008
卷号：XXXVII Part B7
页码：651-656
出版社：Copernicus Publications
摘要：The woodland which covers mainly the district in Jinyunshan Mountain ,Zhongliangshan Moutain, Tongluoshan Mountain and Mingyueshan Mountain area , is very important to accommodate ecological environment ，and is called the "lung" of Chongqing . But these limited resource is dieing out gradually in the process of urbanization and industrialization .In order to keep them from disappearing, we must make certain the distributing of the woodland and the impervious surface, setup corresponding control zone, and restrict human activities.This paper discusses about the extraction of woodland and impervious surface from spot5 images using the Object-oriented classification method. The high-resolution satellite images provide more detailed spatial information such as texture，shape，size and context rather than spectral information. Classification relies on the pixel-based approaches is limited at present。Typically，they have considerable difficulties in dealing with the rich information content of high-resolution data，they produce inconsistent classification results and "Salt and Pepper" , and they are far beyond the expectations in extracting the object of interest. While classification based on object-oriented can overcome these difficulties, and they produce results that could be understood more easily.This paper makes use of the object-oriented approach to the classification of high- resolution imagery, involves the segmentation of image data into objects at multiple scale levels. Class rules are generated using spectral signatures, shape and contextual relationships, and then used as a basis for the fuzzy classification of the imagery. The utilization of spectral, textural, shape properties and fuzzy thinking may reduce the uncertainty in the process of classification. The paper established the vegetation fraction based on the dimidiate pixel model to extract the woodland that need to be protected through investigation on the spot .These is little information in the shadow, so the paper extracts the shadow specially, then using class-related features to determine objects' belongings in the shadow .When it refers to impervious surface, the paper firstly constructs several classes to extract different buildings, then incorporate all of them to make the result more precisely. At the same time, combining with terrain and physiognomy analysis, and other thematic information, nature reserve, water conservation district and scenic area for example ,the paper ascertains the control district. The results show better overall accuracy of the object-oriented classification. The conclusion indicates that object- oriented analysis has great potential for extracting land cover information from high satellite imagery ,and can improve the level of automatization and intelligentization of high remote sensing data process and application
