Buildings Extraction Based On Scattering Features For Polarimetric SAR Images

With the rapid development of synthetic aperture radar (SAR) systems, hugeamount of high resolution, multi-band, polarimetric SAR (PolSAR) and polarimetricinterferometry SAR (PolInSAR) data have been obtained, which make details ofbuildings information in SAR images much more abundant and comprehensive. FullPolSAR image contains more buildings details and polarimetric characteristics, whichmakes it more accurate for buildings detection and outlines extraction. And theinstability of building detection and extraction increased, which presented newchallenges for PolSAR buildings extraction. According to characteristics of domesticairborne polarimetric SAR image, this paper focuses on how to detect buildings andextract building outlines using the frame of different polarimetric scatteringmechanism of buildings with different alignments in PolSAR image. This paper aimsto propose an incoherent decomposition method based on physical scattering modelsuitable for different alignment buildings, and the key points are to study on urban andrural residents building extraction methods: urban buildings detection methods takingbuilding alignments and terrain slope into account is proposed; rural residentialdetection method considering polarization features, texture features, geometrystructure and polarimetric interferometry features comprehensively is proposed. Onthis basis, building outlines can be extracted based on marked watershed segmentationand local Radon transformation. The main work and achievements of this dissertationare as follows:(1) A method to establish buildings optimal features set from SAR images basedon expansion Bhattacharyya distance and modified least angle regression (LARS)algorithm was proposed, which solves the problem of the complex feature separatedinto amplitude and phase. According to the characteristics of domestic airbornePolSAR/PolInSAR image, the texture, geometry, polarization, polarimetricinterferometric characteristics were firstly analyzed. The key point is to discusspolarization scattering mechanism of buildings with different alignments. Buildingsoptimal features set was established based on expansion Bhattacharyya distance andmodified LARS algorithm from PolSAR image. The method utilizes the polarizationand the polarimetric interferometry complex characteristics effectively and maintainsthe phase information of the complex features sufficiently. The distinguishingcapacity between buildings and non-buildings with complex features can be evaluated much more accurately.(2) A incoherent decomposition method based on physical scattering modelsuitable for buildings on different aligments was proposed, and it solves the problemof existing decomposition methods, which overestimate the volume scattering power,occurrence of negative power pixels in surface scattering and double-bouncescattering, and oriented buildings misclassified as volume scattering, such asvegetations. According to the deficients of existing methods, the original coherencematrix is compensated by polarization orientation angle, and then phase rotation of thecompensated coherence matrix is introduced. So the unitary transformation forcomplex form of coherent matrix can be realized, and the phase rotation can reducethe influence of terrain slope and oriented building; a extended volume scatteringmodel fit for oriented buildings is proposed, and it contains pure scattering model andoriented dihedral scattering model which can distinguish oriented buildings fromvegetations; and finally, two energy constraints are added and it can avoid negativepower pixels effectively from surface scattering and double-bounce scattering.(3) Urban buildings with different alignments and rural residential areasdetection methods were proposed based on polarization scattering characteristics ofbuildings, which solves the problem of detecting buildings with an angle betweenradar azimuth and buildings alignments or on slope terrain. Urban buildings detectionmethod based on improved four-component decomposition was proposed. Andnormalization circular polarization correlation coefficient was introduced into theoriented buildings detection. It takes buildings alignments and terrain slope intoaccount and preserves the buildings polarimetric scattering characteristics. Themethod corrects the problem of misclassified buildings scattering mechanism.Considering hash and disordered highlights characters for rural residential areas inpolarimetric SAR image, multi-features fusion detection method was presented. Themethod combines features of the optimal features set and the improvedfour-component decomposition based on polarimetric scattering characteristics ofbuildings, which improves the precision of detecting rural residential areas.(4) A hierarchical frame which contains the buildings feature analysis, optimalfeature set construction, buildings detection and buildings outlines extraction has beenset up. Based on the results of buildings detection, building outlines extractionmethods based on marked watershed segmentation algorithm and local Radontransformation were studied respectively and building outlines can be extracted effectively.