Building Extraction From PolSAR Imagery Based On Polarimetric Information And Improved FNEA Segamentation

Polarimetric SAR is a new type of radar, which can receive the scattering information of different polarized scattering channels. The appearance of PolSAR has greatly enriched the interpretation of SAR data information. Polarization information in different polarization channels reflects the response of different types of ground objects to different polarization radar waves, and response of buildings in different polarization channel can provide a new perspective which is different from optical information to guide the building extraction. However, at present, most of the researches on building with polarization information were based on pixel level, less research on the object oriented.In this thesis, the key points of object-oriented building extraction were segmentation and extraction. Firstly, when FNEA was applied to PolSAR images, the edge of result shifted. So the edge feature was introduced to guide the pixel merging to improve this problem. Then, the method of building extraction based on polarization scattering characteristics made use of a single feature, which made the extraction accuracy of buildings poor. Thus, polarization orientation angle characteristics, brightness features and texture features were utilized to improve accuracy. And through three parallel experiment in the airborne polarimetric data, the effectiveness of the improved method of segmentation and building extraction was proved. Details were as follows:(1) The polarization scattering characteristics of the target were summarized based on Jones vector, polarization ellipse and Stokes vector, and several speckle filtering methods for polarimetric SAR images were compared .(2) A segmentation method of PolSAR image based on improved FNEA was proposed. FNEA's result showed edge shift because of the lack of edge information in PolSAR image. So that, the edge feature were used to guide pixel level merging to generate initial objects. Based on that, multiscale segmentation results were completed after the calculation of the generalized similarity and the object merging.And the optimal scale segmentation result was obtained with the single-feature building extraction accuracy as the criterion,which was used to extract the object-level building subsequently.(3)A building extraction method based on multiple features for PolSAR images was proposed. The existing methods used a single feature, polarization decomposition characteristics, resulting in missed detection and fault detection in building extraction. To overcome this problem, the reason of missing detection and fault detection and the polarimetric characterization of that were analyzed, and polarization orientation angle, brightness and texture features were taken into account. The specific steps of the method and the key problems in the implementation process are discussed in detail. At the same time, the object dominance factor is proposed to improve the traditional scattering dominant method.To sum up, an object-oriented synthesis of multi-feature PolSAR building extraction method is formed...