INCOHERENT TARGET DECOMPOSITION OF POLSAR DATA FOR LAND APPLICATIONS

Abstract

The rapid growth of population with limited natural resources demands a study of anthropogenic modification of Earth surface and its environmental impact. Earth surface classification, popularly known as land-cover classification, is the first step to achieve this goal. Land-cover alteration is strongly associated with atmosphere, ecosystem process and human behavior, thereby can be related to worldwide climate change. The manual collection of Earth surface parameters is a herculean task. Also, the dynamic changing environment of Earth demands frequent observations of its surface. The most adequate alternative which provides the feasible solution to the challenge of timely observations of Earth surface in all weather conditions is polarimetric synthetic aperture radar (PolSAR) system. Since the backscattered polarization information depends on the sensitivity of the transmitted polarization to the dielectric and geometrical characteristics of scatterers, it is easier to extract the geophysical and biophysical parameters of Earth’s surface with the fully polarimetric data. The PolSAR observed values are transformed into the parameters associated with the physical significance of the scatterer by so-called target decomposition theorems. Target decomposition theorems exploit the PolSAR information to interpret the scattering mechanisms. Since different types of land-covers involve different types of scattering, correct interpretation of the underlying scattering mechanism is the foundation of land-cover classification. A straightforward way to accomplish this is through the decomposition of PolSAR data into linear sum of physical scattering mechanisms. This principle is popularly referred to as model-based decomposition. Although model-based decomposition technique for PolSAR data has been investigated for the last two decades, efficient and robust methods are still very few in number. The reason behind this is the inherited flaws of model-based decomposition technique. The efficacy of a model-based decomposition technique is mainly based on the appropriate modeling of physical scattering mechanisms. However, the scatterings from terrain targets highly depend on their relative orientation with respect to the radar illumination. Thereby, different type of scatterers with different orientations may generate similar scattering and vice versa, which give rise to scattering mechanism ambiguity. If the physical scattering mechanisms are not properly modeled, scattering mechanism ambiguity may lead to misclassification of land-covers. Another consequence of scattering mechanism ambiguity is the overestimation of some of the scattering powers. This overestimation may result in negative scattering contributions for other scattering mechanisms. The occurrence of negative scattering powers indicates that the backscattered polarimetric information is not properly modeled. The third issue related to the modelbased methods is the under-determined equation system. Due to fewer number equations than unknowns, the branching conditions have to be applied in model-based decomposii tion methods. Branching conditions are the assumptions or constraints taken to solve the under-determined equation system of model-based decomposition methods. Because of this, the performances of model-based decomposition methods depend on the efficiency of branching conditions. These limitations restrict the applications of model-based decomposition methods. In order to resolve these issues, some feasible solutions are presented in this thesis. The research work of this thesis can be divided into two main parts. In the first part, straightforward solutions to the challenges regarding scattering mechanism ambiguity and negative power problem of model-based decomposition methods are presented. The first part can be further broken into three sub-research works. In the first research work, a new urban scattering model is presented to deal with scattering mechanism ambiguity between vegetation and oriented urban areas. In the second work, unitary matrix rotations are applied to decouple the energy between co- and cross-polarization scattering mechanisms. This decoupling optimizes the PolSAR coherency matrix to be used for model-based decomposition methods. In the third research work, hybrid scattering models are utilized to address the negative power problem of model-based decomposition methods. The second part of the thesis investigates the significance of branching conditions in model-based decomposition methods. The first section of this second part presents computationally efficient alternate model-based decomposition schemes. These alternate decomposition schemes demonstrate a methodology to linearly solve the under-determined equation system of model-based decomposition methods without incorporation of branching conditions. In the second section, an efficient branching condition is presented to enhance the decomposition results of current model-based decomposition methods. In summary, this thesis contributes towards the development of new efficient model-based decomposition methods for land-cover classification. Simple approaches are also explored to optimize the PolSAR coherency matrix and to enhance the performance of existing model-based decomposition methods.