Research And Application On Polarimetric SAR Incoherent Target Decomposition

As a milestone-like breakthrough of quantitative measurement tool in high-resolution microwave imaging system,Polarimetric Synthetic Aperture Radar(Pol SAR)can completely and accurately acquire scattering information from different targets by measuring the amplitudes and phases under different polarimetric combinations.Between the input data and real applications,due to its explicit physical significance,polarimetric incoherent target decomposition(PITD)plays an irreplaceable role in scattering mechanism interpretation without any prior knowledge.Nowadays,urban development has become an important symbol of the progress of human civilization in the new era and therefore,building characteristic connotations derived from building damage assessment,road planning,and hazard change detection require scientific evaluations.In view of the demand of strategic intelligence acquisition,disaster monitoring,and government public decision-making from military and local authorities,this thesis systematically investigates the model-based PITD where the focus is scattering behavior analysis,multi-mode scattering modeling,and buiding information extraction.The main content can be summarized as follows:(1)Scattering power tranferation-based PITD.The classical Freeman three-component and Yamaguchi four-component model-based decompositions are firstly revisited.According to the deficiencies of the involved scattering models and decomposition schemes,the scattering ambiguity between obliquely oriented buildings(OOBs)and vegetated areas which caused by the overestimation of volume scattering(OVS)is analyzed.In order to reduce the cross-pol powers in OOBs,a hierarchical and generalized model unitary transformation-driven PITD is proposed.First,the proposed method implements two unitary similarity transformations on the input matrix and scattering models by orientation angle compensation and phase angle compensation.On the basis that polarimetric information remains unchanged and maximum-utilized,the proposed method then incorporates the generalized volume scattering model and cross scattering model to decompose the input matrix hierarchically according to the ratio of polarimetric coherence.Experiments conducted on real Pol SAR data verify the effectiveness in improving the OVS and eliminating the scattering ambiguity of the proposed method.Moreover,different comparative experiments indicate that the phase angle transformation can better reduce the cross-pol powers and constrains the occurance of negative scattering powers.(2)Scattering orientation extension-based PITD.First,the scatteing behaviors of buildings(especially OOBs)are analyzed and the impacts of building orientations on scattering mechanism interpretation and scattering modeling are discussed.According to the building geometry,we point out that the polarimetric orientation angle(OAC)of buildings can be extended to the second dimensionality in fact.On this basis,a neighborhood-adaptive arc distance median filtering algorithm and a slope-induced shape-from-shading technique are proposed to modify the first dimensional POA and estimate the second dimensional POA,respectively.To incorporate the POA information into the scattering mechanism interpretation,a more generalized scattering model called the doubled cross scattering model is constructed,thus presenting a refined hierarchical model orientation extension-driven PITD.The hierarchy reflects in that the decomposition is implemented on two stages based on the proposed element-driven building scattering feature.Qualitative and quantitative experimental results confirm the existent rationality and effectiveness of the second dimensional POA.Refined decomposition results indicate that the doubled cross scattering model can not only moderate the scattering ambiguity but also characterize the cross-pol component effectively.(3)Scattering component distribution-based PITD.The classical Cloude decomposition and the eigenvalue-based parameters,i.e.,polairmetric entropy,averaged scattering angle,and polarimetric anisotropy are first reviewed.According to the interferences of coherent noise and multilook processing for eigenvalue parameter estimation,the deficiencies of classical eigenvalue-based parameters in land cover classification are discussed.To address the interference issue,two equivalent emerging eigenvalue-based parameters,namely,radar vegetation index and polarimetric asymmetry are introduced to construct the OOB scattering feature descriptor in a qualitative manner.Based on the fact that the cross-pol component is significantly larger than the co-pol component in OOBs,a more pragmatic scattering model called the OOB scattering model is proposed by combining the cross/doubled cross scattering model and the OOB scattering feature descriptor.Accordingly,a generalized model feature modification-driven five-component PITD is proposed.Experimental results demonstrate that the OOB scattering model can reasonably distribute the total cross-pol component without any compensation or hierarchy.Moreover,the OVS is remarkably moderated and more reasonable surface scattering is enhanced.More importantly,the OOB scattering components are further enhanced,which enable better characterization of building scattering.(4)PITD-based Pol SAR building information extraction.First,researches on building detection are carried on.On the one hand,according to the fact that the double-bounce co-pol component is widespread in buildings approximately aligned with the flight trajectory,a change detection-based power saliency detection method is proposed.On the other hand,through analyzing the scattering randomness,polarimetric asymmetry,and depolarization effect of OOBs,an eigenvalue-based scattering saliency detection method is presented.Experiments conducted on real Pol SAR data indicate that the fusion of these two preliminary detection results not only can remove the natural false alarms but also can further improve the detection accuracy of buildings with different orientations.Second,researches on building edge extraction are carried on.Genenally,traditional bilateral windows gather inconsistent pixels due to their fixed shapes and sizes.To overcome this drawback,a scattering mechanism-driven adaptive window which has flexible shapes and sizes is generated based on the hierarchical model-based decomposition.Moreover,in order to circumvent the risk of making statistical distribution assumption,an optimal contrast ratio-based measurement is further incorporated.Experiments conducted on real Pol SAR data demonstrate that the intergration of scattering charactistics and optimal contrast can achieve higher edge location accuracy and preserve more local details.Third,researches on building segmenation are carried on.To address the issues that the relationship between the simple linear iterative clustering and global image information is not clear and the computational complexity is extremely high in the normalized cuts,a linear feature clustering superpixel segmenation method is proposed through constructing a higher dimensional feature space defined by a positive semi-definite kernel function.In the clustering,the extracted edge and other homogeneity information is adaptively considered for each pixel.Moreover,the proposed method adopts features derived from the hierarchical model-based decomposition to overcome the segmentation ambiguity between OOB and natural area pixels.Theoretical and observational evidences show that the proposed method generates visually pleasing superpixels with favorable boundary adherence and local information perseverance.Moreover,the segmentation ambiguities are commendably restrained...