| Περίληψη: | Radar image analysis is the cutting edge technology in Remote Sensing for observing the Earth’s surface. Radars as passive sensors acquire the backscattered solar illumination from the Earth’s surface and as such they are subject to atmospheric conditions. On the other hand, radars can be active sensors that transmit polarized electromagnetic waves towards the Earth and then receive the echoes that are backscattered. Such an active radar system is independent of solar illumination, allowing day and night imaging. Nowadays radar systems are capable of providing very high resolution images of the Earth’s surface by means of Synthetic Aperture Radar (SAR) systems. SAR imaging is ideal to monitor dynamic processes on the Earth’s surface in a reliable, continuous and global way. A SAR system operates in the microwave region of the electromagnetic spectrum making it, also, independent of any weather effects, smoke, fog and other related phenomena. Given the potential of such high resolution images, SAR imaging constitute a vast and active research area with applications in landslides observation, earthquake monitoring, maritime security, search and rescue missions, automatic target recognition and many more.
A modern SAR system can transmit electromagnetic waves in two different polarizations, namely horizontal and vertical, and can receive their backscattered echoes in the same two polarizations. In this way, four combinations of transmitting and receiving signals are created. Each of the transmitting-receiving signal combination constitutes a single image of the same geographical scene, while the four images altogether constitute a fully-polarimetric image. The fully-polarimetric image expresses the polarimetric properties of the Earth’s scene that is depicted on it. In this Ph.D. thesis the polarimetric nature of targets that are depicted in a SAR image has been extensively studied. In this context, a target is considered as a single or a group of pixels that form a unique and distinctive object while a complex target is considered the one that has a complex physical structure, such as ships, buildings, plants and etc.
Radar polarimetry studies the way in which a radar signal interacts with a real target aiming to deduce its physical and geometric properties. It is long known that when a transmitted electromagnetic wave is being scattered from a target its polarimetric properties are being subject to changes. These changes are directly linked to the scatterer physical and electrical properties. Several researchers such as J. Huynen, E. Luneburg, S. R. Cloude, W.L. Cameron, R. Touzi and more have advanced the theoretical framework of polarimetric decompositions, theories that constitute the extraction of meaningful physical properties from a target. Polarimetric approaches are mostly divided into two categories, in Coherent Target Decompositions (CTDs) and in Incoherent Target Decompositions (ICTDs). Each approach presents its own advantages and disadvantages with respect to scatterer representation and classification to elemental scattering mechanisms. However, it is observed that, under certain conditions, different CTDs assign different physical properties to a single scatterer which seems rather irrational. The physical properties of a scatterer cannot change with respect to different representation models. The classification of scatterer by means of a CTD must always be the same, regardless the additional information that a different representation model may provide. In this thesis the polarimetric nature of the elemental scatterers, the relationship between the two most famous CTDs, the Huynen- and Cameron CTD, is examined and a unifying approach is presented.
Additionally, this study is focused on the evaluation of the interferometric information of a SAR system from a polarimetric point of view. Interferometric SAR (InSAR) imaging has the means for generating digital elevation models (DEMs) and ground deformation maps that present the relative deformation of the Earth’s surface between acquisitions. Differential Ιnterferometry (DInSAR) is a technique derived from classic Interferometry in order to produce interferograms from which the topographic contribution has been eliminated. Even so, DInSAR inherits many difficulties of InSAR with the main being the temporal and spatial decorrelation. However, single pixels that are coherent over long time intervals and over wide look-angle variations are able to overcome the aforementioned obstacles. In this thesis the polarimetric nature of the so-called persistent scatterers is evaluated and their use as an auxiliary tool for preprocessing a scene is assessed.
Furthermore, a novel CFAR ship detector that exploits the scattering mechanism of a scatterer instead of its amplitude information is developed. Ships are depicted by pixels that present strong echoes on a background with relatively low backscattering power formed by the surface of the sea, making ship detection feasible. The vast majority of the literature regards the development of CFAR ship detectors using only the backscattering amplitude information from single polarimetric images. Under this framework, one must take under consideration several factors regarding sea-ship distinction with the sea-states, the local weather phenomena and the combination of sea and Kelvin wakes being of the highest importance. However, these detection approaches would fail to perform when the radar signals present backscattering power of the same order. Therefore, the polarimetric nature of the radar signals must also be examined. That being said, there has been made only a small effort on fully-polarimetric ship detection and there is still a significant lack of a relative systematic and automatic experimental work. In this thesis our aim is to connect polarimetric decompositions with stochastic methods to achieve an automatic and systematic ship detection procedure based on the geometric properties of the scene scatterers. A novel ship detection scheme based on first order Markov models along with Cameron’s decomposition elemental scatterers is introduced. Cameron’s symmetric scatterers will be used as the quantized structural elements of the scene. Nonetheless, the scene scatterers distribution is owed solely to the local topography of the scene and the random position of the ships. Thus, in order to overcome the stochastic nature of the composition of the scene the first order Markov chain model is utilized. First-order Markov chains are adequate to describe the stochastic nature of the polarimetric properties of a scene making the detection procedure feasible. In this way, the dependence of the CFAR detectors to power of the radar signals is overcome. Achieved ship detection performance is high for sea monitoring.
|
|---|
