The Inconsistency Of Polarimetric SAR Model-Based Target Decomposition

Synthetic Aperture Radar(SAR) is paid more and more attention with its capacity of penetrating cloud, fog, dust and gaining surface information in large-scale regions. As the most advanced SAR system, PolarimetricSAR(PolSAR) contains more abundant information than single polarimetric SAR and can uncover and describe the objects inner scattering mechanism. Owing to its exclusive imaging, the scattering type of each pixel, in fact, can be represented as the sum of various ones. The incoherent model-based target decomposition developed based on Freeman decomposition can be described as the linear-weighted sum of some scattering types. It can also represent the scattering process more intuitionally and is playing an important role in the physical parameters analysis or inversion, which makes it the hot topic at present. However, the inconsistency problem is lying in it, showing that the power of the decomposed components is negative, which hinders the objects analysis and interpretation. This thesis is based on the adequate understanding about the PolSAR model-based decomposition to explore the inconsistency problems in three-component and four-component model-based decompositions in detail respectively. Meanwhile, some algorithms have been developed to solve the problem, aiming at solving the inconsistency problem. Lastly, the decomposed features combing with other polarimetric features are applied to the buildings segmentation. Therefore, the research we have done can be described as follows:1. The issue of the inconsistence has been induced and analyzed in detail based on covariance matrix for three-component decomposition. In order not only to address this problem, but also to solve the problem existing in urban areas after normalization process, a three-component decomposition model based on de-orientation and a generalized volume scattering model is developed. Firstly, de-orientation is applied to the covariance matrix to reduce the volume scattering power before it is decomposed into three scattering components. Secondly, a Generalized Volume Scattering Model (GVSM) is adopted by considering the change of HH and VV ratio in different forest areas. In addition, power constrain method is added to eliminate negative power completely. The results are validated by using the L band E-SAR data in Oberpfaffenhofen area in Germany and comparing with many other methods. The results show that the integrated model can solve the volume scattering overestimation more notable than other decomposition model, and the results are closer to real scattering type.2. The issue of the inconsistence has been induced and analyzed in detail based on covariance matrix for four-component decomposition. In order to address this problem, we proposed an integrated four-component model-based decomposition based on multi-look covariance matrix. Unlike other improved four-component decompositions that de-orientate the entire data and introduce the helix component into all natural areas. This decomposition can effectively suppress the helix components which are difficult to be produced in most natural areas. This decomposition integrates the selective de-orientate and the generalized volume scattering. Firstly, a new complex Mixed Polarization Correlation Coefficient (MPCC) is proposed which not only can be used for selective de-orientation but also can be a criterion to judge whether the target is reflection symmetric. Then, a generalized volume scattering model is still adopted being the same with that used in three-component model-based decomposition. Finally, for purpose of eliminating negative power pixels completely, based on the normalization method, another power constrain procedures is added which can partially embrace an incoherent decomposition being similar to Krogager coherent decomposition. The effectiveness of the IFMD is demonstrated by the L band airborne E-SAR data, and its applicability for the short band is also discussed using X band spaceborne TerraSAR data.3. It is well known that the existing segmentation methods mostly directly use backscatter coefficient images or span images for segmentation, or a certain kind of polarization characteristic images, and do not make full use of the polarimetric characteristics and spatial characteristics. We propose a buildings segmentation model based on PolSAR images in this article, which makes full use of the polarimetric characteristics and spatial characteristics of buildings by Fractal Network Evolutionary Algorithm (FNEA) and Multiple Linear Regression (MLR) model. By using the proposed model to achieve the segmentation of the L-band EMISAR airborne images in Foulum area and comparing with the traditional methods, It proves that the segmentation model is feasible.