SAR(Synthetic Aperture Radar) is a kind of active microwave sensor. Due to its capacity of all-time, all-weather and high resolution, it has been widely used in many fields such as economy, military and scientific research. At present, the interpretation based on high resolution SAR image data over urban areas has become a research hotspot. In this paper,in-depth researches on building detection from high resolution SAR images and building geometric information extraction have been done. The main research contents are as follows:1. Building extraction from high resolution SAR images. In SAR images, two key features of buildings are layover and shadow. Since the existing methods of building detection based on layover and shadow ignore spatial relation, the extraction results cannot ensure accuracy. In order to improve the effect of neighborhood systems on energy item, the adaptive weight coefficients are introduced to improve MRF(Markov Random Field) in this paper. The improved MRF model is used to classify SAR images. On the basis of classification, the buildings are extracted from SAR images by post-processing, which is based on shape characteristics of buildings such as shape index, aspect ratio and area factor. The efficiency and accuracy of building detection based on improved MRF have been verified through experiments.2. Building geometric information extraction. On the basis of analyzing the defect of direct measurement, the method to obtain geometrical information is proposed via matching the geometrical model of a building and the real SAR image. At first, buildings are detected from SAR images. Then, the matching functions based on the geometrical model of a building and the real SAR image is designed. At the same time, the geometry information extraction is transformed into matching function optimization problem. Given the advantages of genetic algorithm and simulated annealing algorithm, genetic simulated annealing algorithm is applied to optimize the matching function. It has been proved through experiments that the proposed algorithm can obtain more accurate geometry information. |