Font Size: a A A

Classification And Polarization Azimuth Compensation Of Full Polarization SAR Image

Posted on:2015-03-09Degree:MasterType:Thesis
Country:ChinaCandidate:J Y RenFull Text:PDF
GTID:2298330422976231Subject:Computational Mathematics
Abstract/Summary:PDF Full Text Request
As an advanced means of remote sensing information acquisition, fullpolarization SAR (Synthetic Aperture Radar) records more completescattering information of targets.The full polarization SAR classificationfigure can not only provide auxiliary information for target detection or edgeextraction, it also can be used as the final result. Compared with ordinaryremote sensing images, full polarization SAR classification has more researchvalue for revealing polarization scattering information.In order to improve the classification accuracy, full polarization SARimage classification methods are studied in this paper.Currently there are two main methods doing the full polarization SARground object classification: unsupervised classification and supervisedclassification. The advantage of most unsupervised classification methods isproviding auxiliary information for assigning the finally terrain types. Butduring the process of analyzing large-scale remote sensing data we must relyon human experts to participate in the decoder because each clustercorresponds to only a single scattering mechanism and it does not representthe actual terrain. Based on pixels or small region, the supervisedclassification methods with the underlying characteristics is very effectivewhen classifing the ground object which has a single scattering mechanism,but is trouble for complex terrain. According to the two problems, this paper proposes a supervised classification method using a middle layer feature ofMLF (Middle-Level-Feature). The MLF of a pixel is calculated by countingthe frequency of the middle-components in a feature supporting regioncentered on the pixel. Here the middle-components refer to the unsupervisedclustering categories obtained from the low-level polarization characteristics.Then the support vector machine is used for supervised classification after theMLF of all the pixels are calculated. The proposed method is tested on aRadarsat-2full polarization data covering WUHAN area and goodclassification performance and potential of further improvement are shown.The comparison with the supervised classification method combining SVM(Support Vector Machine)and the classic polarization characteristics is given.Different methods for getting the middle-components and feature supportingwindows with different size are studied on their impact on the finalclassification performance.During the much information of the polarization SAR data, thepolarization azimuth angle which is also the azimuth slope reflects the rotationangle of the scattering target relative to the radar line of sight. In polarizationSAR image classification, the two same categories may be divided intodifferent categories just because of the different Orientation to the slope whichreflecting different polarization characteristics in SAR data. In order toeliminate the error classification and improve the classification results causedby terrain factors, this article has carried on the compensation of polarization Angle.Based on DEM (Digital Elevation Model),this paper estimate thepolarization azimuth angle and do the azimuth angle compensation. theexperiment indicated that a part of pixels’ all three scattering components inFreeman decomposition were decreased after the polarization Anglecompensation,but the different is enough small to ignore it. Considering ofCloude decomposition, the ability of distinguishing the two weak scatteringcomponent is enhanced and the randomness of scattering medium is enhanced,at the same time about60%of the pixels’ alpha value which is on behalf ofthe scattering process physical mechanism decreases after polarizationazimuth Angle compensation. With the conflict between Cloudedecomposition which has the characteristics of rotation invariance, it alsoneed further research.
Keywords/Search Tags:Full Polarization SAR, Middle-Level-Feature, Middle-Components, Polarization Azimuth Angle Compensation
PDF Full Text Request
Related items