Electromagnetic Modeling andInformation Extraction from SAR Data

Abstract

The key goal of this thesis work lies in the development of models and tools in support of value-added information extraction from Synthetic Aperture Radar amplitude-only images. In the last decades earth observation instruments provided a great amount of images relevant to any part of the world. These data could be potentially helpful for a wide range of human activities, ranging from agriculture to rural and urban planning and disaster monitoring and assessment. However, practical use of these data is often limited by the lack of efficient, possibly unsupervised, tools for the retrieving of effective information. In this thesis the first steps toward a modeling of the whole imaging process is provided. In particular, we discuss in detail the fundamentals of the Synthetic Aperture Radar in its standard and well known working configuration, highlighting the need for an adequate modeling able to guarantee effective high resolution data description (Chapter 1). In fact, the statistics of this kind of images are often very different from those used in the modeling of low resolution data. First results coming from the analysis of the first TerraSAR-X high resolution data are presented here and represent the first original contribution of this thesis. In Chapter 1 not only the working geometries and SAR performances are presented but also a conceptual scheme for the simulation of the primary signal collected by the sensor called raw data. Simulators, in fact, are important tools supporting the design and project of new sensors and are able to conveniently lead the criterions for setting the mission parameters as they take into accounts the applications they are planned for. Furthermore, they can be used to conveniently address the inverse problem starting from the complete solution of the direct one. In fact, the development of effective information extraction techniques from SAR data and the synthesis of automatic tools for image analysis mandatory pass through the development of adequate direct models relating the image to the parameters of the surface. Thus, the direct models can become the starting point toward the availability of inversion techniques and physically-based classification techniques. The models used in this thesis work are detailed in Chapter 2. In particular, the geometric and electromagnetic models for natural surfaces are presented, both for natural terrain and for the ocean sea surface. After having introduced the different techniques to collect and model SAR data, we move to discuss the possibility of retrieving information analyzing those data. In particular, in Chapter 3 we present a fractal framework for the simulation of SAR images relevant to simulated disaster scenarios. Such an instrument can be used to increase the understanding of the physical mechanisms underlying radar image formation in case of disasters. In, fact, the main problem of the scientist working on the development of remote sensing techniques for disaster monitoring is the lack or the limitedness of an accurate ground truth. The proposed simulator makes possible to perform parametric studies on canonical disasters scenarios with a perfectly known ground truth. Furthermore, it can be used to obtain images relevant to both pre- and post-crisis situations, providing the possibility to develop a test bed of simulated images to be used for the testing of change detection techniques. Relevant case studies are presented with regard to different kinds of natural disasters. Finally, a novel change detection technique based on the estimation of significant parameters and supported by fractal concepts is described. Results on the simulation of images relevant to ocean scenes covered with oil slicks of arbitrary shape are also presented. In Chapter 4 we cope with the problem of radar imaging of fractal surfaces. In particular, we develop a rigorous analytical formulation for the problem in case a small slope regime can be assumed for the profile. The proposed model allows for the computation of the structure function and of the power density spectrum of the image in closed form. The proposed model is validated through an appropriate numerical framework base on the sound physical models presented in Chapter 2. The first steps toward the extension to the two-dimensional case are also provided. Note that the development of this kind of direct modeling is of key importance for every image analysis technique based on the evaluation of global statistics on SAR images.