Land Coverage Classification Based On Spatial And Radiation Characteristics In Hr SAR Image And System Design

With the continuous development of the synthetic aperture radar (SAR) technology, the image sensor can provide great amount of image which is all-weather conditions, high-resolution, detailed ground mapping information. However, faced with a large scene of SAR image, totally dependent on human interpretation takes a lot of manpower, which can not adapt to real-time requirement of military and emergency response. In addition, because of the presence of speckle noise and the special SAR imaging mechanism, it result in loss of structural features, which is most different from the optical image in visual effect, the scene is difficult to understand, even to rely on manual to understand, it still very difficult, and it has the existence of artificial subjectivity. If these difficulties are not addressed, it will directly affect the real application of the high-resolution, large scenes, high-quality SAR images.According to China's status and the development of the SAR system and the application needs in SAR image land coverage (LC) classification. This paper focuses on the core technology of high-resolution SAR image LC classification, including the LC coarse classification technology based on radiation characteristics and spatial information, the optimized technology bases on high-level language and neighborhood information, the land coverage precise classification technology based on GIS information and topology information. Finally, we introduce the design and implementation of the LC classification system for the high-resolution SAR image.The LC classification technology in this paper focuses on the scattering characteristics and spatial correlation characteristics of the typical scenes in SAR images, and the specific content is as follow: First, according to the basic characteristics of SAR image, we extract reliable features of the typical scene to divide the image into built-up areas, forests, open land areas and water; Then, according to high-level semantic information and GIS geographic database, we optimize the LC coarse classification results; Finally, according to the spatial relationship features and DEM data, we refine the LC coarse classification results from the four typical scenes to ten scenes which include urban area, rural area, mountain forest, plain forest, farmland, desert, flat areas, rivers, lakes, and oceans.The LC coarse classification technology based on radiation characteristics and spatial information is the basis of the SAR LC classification system. This paper analyzes the inherent characteristics of SAR image, and study the scattering properties and spatial relations of the typical surfaces in SAR image, and then find the best classification strategy according to these features. To ensure the stability of this classification algorithm, this paper introduces the concept of false texture, to estimate the extent of SAR image affected by noise, and to use these features to design a stable classification algorithm, which has a better classification result for the high-resolution SAR image.In order to optimize the results of coarse classification to improve LC classification accuracy, this paper introduces the optimized technology bases on high-level language and neighborhood information. The paper mainly discussed the most decision-making technology, the forest and built-up areas identification technology based on context and the water and the shadow identification technology based on radar imaging priori knowledge, which helps enhance the performance and robustness of LC classification system.The land coverage precise classification based on GIS information and topology information is for the refinement of LC coarse classification result. According to the characteristics of build-up areas, forests, waters and open land areas, combined with GIS information and topological shape information, this paper is to propose the refined algorithm to extract the refined information, and establish a model of LC refined extraction.The land coverage classification algorithm of high-resolution SAR image is a summary of the master's research. The author hopes that it can solve parts of the problems of the land coverage classification of SAR image. There are many places to be improved and studied in-depth. And the ever-changing technology and needs of remote sensing technology, which all require more extensive and in-depth research.