Underlying Topography Inversion Using TomoSAR Based On Covariance Matrix Optimization

Forests,as the largest terrestrial ecosystem on Earth,are the origin and habitat of human beings and many animals and plants.They have an irreplaceable position in the ecosystem.Modeling their ecosystem and surveying their important resources are urgent and challenging research topics nowadays.The underlying topography,as an important parameter for forest resource survey,is also an essential part of the global digital terrain model.It not only affects the spatial distribution of forest resources,but also closely relates to the stability of forest ecosystem.Synthetic aperture radar tomography(Tomo SAR)technology,developed over the past two decades,is a new three-dimensional(3D)microwave remote sensing technology that compensates for the limitation of traditional two-dimensional(2D)SAR,which can only obtain 2D information of the target.This technology does not require any special processing of the trajectory of the flying platform or control of motion errors,as it relies on multiple imaging of the target ground object to form a new synthetic aperture in the upward direction.By constructing the third dimension with increased resolution,it is possible to reconstruct the 3D structure of the object.Given these advantages,Tomo SAR technology is an excellent choice for forest underlying topography inversion and even 3D structure modeling.For distributed scatterers such as forests,the vertical backscattering power is contained in the magnitude and phase of the covariance matrix,and thus forest SAR tomography 3D imaging always processes the covariance matrix.Most current methods use a local means value of the sample covariance matrix when solving the tomographic function model,which may not guarantee the accuracy of the valuation obtained and is prone to problems such as mixed superposition of different scattering mechanisms,poorly refined spectrum,and loss of detailed information.In addition,under the limitation of the number of spatial baselines and the influence of imaging conditions,the spectrum will inevitably produce more or less sidelobe effects,which greatly affects the interpretation of the spectrum and the subsequent parameter inversion.At the same time,the inversion of underlying topography often uses single-polarization data sets.Although singlepolarization data are easy to acquire,have large swath width and small computational load,they cannot fully capture the reflection characteristics of ground objects.Therefore,increasing observation data to compensate for the lack of ground object information is also an effective way to improve inversion performance.To address the above issues,this thesis focuses on a covariance matrix optimization-based Tomo SAR 3D imaging method,which is developed from two aspects of observation data optimization and data purification of the solved data,while considering the introduction of multi-polarization data to serve the major need of high accuracy in an inversion of forest underlying topography.The main research contents and contributions of this thesis are as follows.(1)A tomographic 3D imaging method for single-polarization SAR based on covariance matrix outside optimization is proposed.This method utilizes a non-local means method to identify neighboring pixels with high similarity to the target pixel,thereby comprehensively reflecting its feature information,reducing the influence caused by irrelevant interference signals,and improving the accuracy of the covariance matrix estimation.To verify the feasibility and effectiveness of the method,this thesis performs SAR tomography imaging using six Bio SAR 2008 L-band fully polarization airborne SAR images collected in the boreal forest in the Krycklan region of northern Sweden,and performs inversion analysis from the profile and the global,respectively.The experimental results show that the non-local means method in this thesis has a more significant improvement over the traditional local means method.The non-local means method can not only better separate the scattering phase centers of the ground and canopy,but also has better inversion performance in places with large variations of topographic relief.In terms of specific data,the inversion accuracy of the non-local means method is improved by more than 30% compared with the local means method.(2)A tomographic 3D imaging for single-polarization SAR method based on covariance matrix inside optimization is proposed.This method reorganizes the acquired covariance matrix,assigns different weights to its eigenvectors,and linearly combines the eigenvalues and eigenvectors of the covariance matrix to enhance the true reflection signal and reduce the spurious interference signal while minimizing the effect of the sidelobe appearing in the spectrum.To verify the feasibility and effectiveness of this new low sidelobe nonparametric spectrum estimation method,validation experiments are carried out using simulated data and six Afri SAR 2016 P-band fully polarization airborne SAR images collected from a tropical forest in the Lope region of Gabon,Africa,and the results show that this method can effectively reduce the sidelobe while improving the accuracy of forest underlying topography estimation.(3)A tomographic 3D imaging method for full-polarization SAR based on joint optimization inside and outside the covariance matrix is proposed.This method utilizes fully polarization data to compensate for the defect of incomplete signal in singlepolarization data,thus achieving the goal of obtaining sufficient original information and accurate feature information extraction.To verify the feasibility and effectiveness of this method,this thesis uses four Tropi SAR 2009 P-band fully polarization airborne SAR images acquired from the tropical forest in the Paracou region of French Guiana to perform the inversion of the forest underlying topography.Compared with other methods,this method demonstrates advantages in global solution with less overestimation or underestimation,resulting in minimal error between its estimated underlying topography and Li DAR underlying topography measurements.