Vis enkel innførsel

dc.contributor.advisorEltoft, Torbjørn
dc.contributor.authorDoulgeris, Anthony Paul
dc.date.accessioned2011-02-24T08:04:56Z
dc.date.available2011-02-24T08:04:56Z
dc.date.issued2011-02-18
dc.description.abstractThis thesis describes general methods to analyse polarimetric synthetic aperture radar images, with the primary application of unsupervised image segmentation. The fundamental physics of electromagnetic scattering from distributed targets leads to the image phenomenon called speckle. Speckle is shown to be potentially non-Gaussian and several statistical distributions are investigated. Non-Gaussianity and polarimetry both hold pertinent information about the target medium and methods that utilise both attributes are developed. Two approaches are proposed: a local feature extraction method; and a model-based clustering algorithm. The local feature extraction approach creates a new six-dimensional description of the image by extending standard polarimetric features with a non-Gaussianity measure. Importantly, the non-Gaussianity measure is model independent and therefore does not unduly constrain the analysis. This may be used for subsequent image analysis or for physical parameter extraction, and unsupervised image segmentation is demonstrated with good results. The model-based approach describes a Bayesian clustering algorithm for the K-Wishart model and incorporates both non-Gaussianity and polarimetry. The initial implementation required several key parameters to be given in advance. When compared to the more common Wishart model, the K-Wishart gives similar results for Gaussian image regions, but performs better for non-Gaussian regions. Further development of the model-based method resulted in a novel technique to automatically determine the number of distinct classes supported by the data, given the model and a statistical confidence level. All relevant parameters are subsequently estimated within the algorithm and no special initialisation is required. These are significant advances from existing methods, where key parameters are predetermined and the number of classes are found after many full clustering results are obtained. The methods are general and apply to all coherent imaging systems that exhibit product model based statistics. The methods are demonstrated on several real radar images and with different numbers of polarimetric channels.en
dc.description.doctoraltypeph.d.en
dc.descriptionPaper number 2 of the thesis is not available in Munin due to publisher's restrictions:<br/>Anthony P. Doulgeris, Stian Normann Anfinsen and Torbjørn Eltoft: ’Classification With a Non-Gaussian Model for PolSAR Data’, IEEE transactions on geoscience and remote sensing (2008), vol. 46, no. 10, pp. 2999-3009 Available at <a href=http://dx.doi.org/10.1109/TGRS.2008.923025> http://dx.doi.org/ 10.1109/TGRS.2008.923025<a/>en
dc.identifier.isbn978-82-8236-022-7
dc.identifier.urihttp://hdl.handle.net/10037/2962
dc.identifier.urnURN:NBN:no-uit_munin_2693
dc.language.isoengen
dc.publisherUniversitetet i Tromsøen
dc.publisherUniversity of Tromsøen
dc.rights.accessRightsopenAccess
dc.subject.courseIDDOKTOR-004en
dc.subjectVDP::Mathematics and natural science: 400::Mathematics: 410::Statistics: 412en
dc.subjectVDP::Matematikk og Naturvitenskap: 400::Matematikk: 410::Statistikk: 412en
dc.subjectVDP::Mathematics and natural science: 400::Physics: 430::Electromagnetism, acoustics, optics: 434en
dc.subjectVDP::Matematikk og Naturvitenskap: 400::Fysikk: 430::Elektromagnetisme, akustikk, optikk: 434en
dc.subjectremote sensingen
dc.subjectearth observationen
dc.subjectsynthetic aperture radaren
dc.subjectpolarimetric radaren
dc.subjectstatistical modellingen
dc.subjectnon-Gaussian statisticsen
dc.subjectclustering algorithmen
dc.subjectimage segmentationen
dc.titleNon-Gaussian statistical analysis of polarimetric synthetic aperture radar imagesen
dc.typeDoctoral thesisen
dc.typeDoktorgradsavhandlingen


Tilhørende fil(er)

Denne innførselen finnes i følgende samling(er)

Vis enkel innførsel