In remote sensing, segmentation is a procedure of partitioning the domain of a remotely sensed dataset into meaningful regions which correspond to different land use and land cover (LULC) classes or part of them. So far, the remotely sensed data segmentation is still one of the most challenging problems addressed by the remote sensing community, partly because of the availability of remotely sensed data from diverse sensors of various platforms with very high spatial resolution (VHSR). Thus, there is a strong motivation to propose a sophisticated data representation that can capture the significant amount of details presented in a VHSR dataset and to search for a more powerful scheme suitable for multiple remotely sensed data segmentations. This thesis focuses on the development of a segmentation framework for multiple VHSR remotely sensed data. The emphases are on VHSR data model and segmentation strategy. Starting with the domain partition of a given remotely sensed dataset, a hierarchical data model characterizing the structures hidden in the dataset locally, regionally and globally is built by three random fields: Markova random field (MRF), strict stationary random field (RF) and label field. After defining prior probability distributions which should capture and characterize general and scene-specific knowledge about model parameters and the contextual structure of accurate segmentations, the Bayesian based segmentation framework, which can lead to algorithmic implementation for multiple remotely sensed data, is developed by integrating both the data model and the prior knowledge. To verify the applicability and effectiveness of the proposed segmentation framework, the segmentation algorithms for different types of remotely sensed data are designed within the proposed segmentation framework. The first application relates to SAR intensity image processing, including segmentation and dark spot detection by marked point process. In the second application, the algorithms for LiDAR point cloud segmentation and building detection are developed. Finally, texture and colour texture segmentation problems are tackled within the segmentation framework. All applications demonstrate that the proposed data model provides efficient representations for hierarchical structures hidden in remotely sensed data and the developed segmentation framework leads to successful data processing algorithms for multiple data and task such as segmentation and object detection.

Synthetic aperture radar (SAR) is one of the most versatile tools ever invented for imaging. Due to its better Rayleigh resolution, SAR imaging provides the highest quality radar imagery. These images are used for many applications including but not limited to terrestrial mapping, disaster reconnaissance, medical imaging and military applications. Imaging techniques or geometries which can improve the resolution of the reconstructed imagery is always desired in the SAR imaging. In this dissertation both the linear and nonlinear frequency modulated chirp signals are discussed. The most widely used frequency modulated chirp signal, linear frequency modulated chirp signal, and some of its properties such as spectrum, point spread function and matched filter are summarized. A new nonlinear frequency modulated chirp signal which can be used to improve the image resolution is introduced. In order to validate the offered chirp signal, spotlight SAR imaging geometry together with 2D polar and Stolt format reconstruction algorithms are considered. The synthetic examples are generated using both chirps both with polar and Stolt format processing. Additionally a new change detection method which depends on the idea of generating two different final change maps of the initial and final images in a sequence is offered. The specific algorithms utilized for testing this method are the widely used correlation coefficient change statistic and the intensity ratio change statistic algorithms. This method together with the algorithms mentioned is first applied to synthetic data generated by Stolt format processing. It is shown that the method works on synthetic data. The method together with the algorithms mentioned is also applied to two case studies dfreal disasters, one is 2010 Gulf of Mexico oil spill and the second is 2008 China Sichuan earthquake. It is shown that two final change map method can reduce the false identifications of the changes. Also it is shown that intensity ratio change statistics is a better tool for identifying the changes due to oil contamination. The data used in this study is acquired by Japanese Aerospace Agency's Advanced Land Observing Satellite (ALOS) through Alaska SAR Facility (ASF), at the University of Alaska, Fairbanks.

The capability to observe ocean swell using spaceborne Synthetic Aperture Radar (SAR) has been demonstrated starting with ERS-1 mission in 1992. This dissertation shows how ocean swell properties can be used to combine swell observations of heterogeneous quality and acquired at various times and locations for the observation and forecast of ocean swell fieldsusing ASAR instrument on-board ENVISAT. The first section is a review of how ocean swell spectra can be derived from the SAR complex images of the ocean surface using a quasi-linear transformation. Then, significant swell heights, peak periods and peak directions from in situ measurements are used to assess the accuracy of the SAR observed swell spectra. Using linear propagation in deep ocean, a new swell field reconstruction methodologyis developed in order to gather SAR swell observations related to the same swell field. Propagated from their generation region, these observations render the spatio-temporal properties of the emanating ocean swell fields. Afterwards, a methodology is developed for the exclusion of outliers taking advantage of the swell field consistency. Also, using the irregularly sampled SAR observations, quality controlled estimations of swell field integral parameters are produced on a regular space-time grid. Validation against in situ measurements reveals the dramatic impact of the density of propagated observations on the integral parameters estimated accuracy. Specifically, this parameter is shown to be very dependent on the satellite orbit. Finally, comparisons with the numerical wave model WAVEWATCH-III prove it could potentially benefit from the SAR swell field estimates for assimilation purposes.

In remote sensing, segmentation is a procedure of partitioning the domain of a remotely sensed dataset into meaningful regions which correspond to different land use and land cover (LULC) classes or part of them. So far, the remotely sensed data segmentation is still one of the most challenging problems addressed by the remote sensing community, partly because of the availability of remotely sensed data from diverse sensors of various platforms with very high spatial resolution (VHSR). Thus, there is a strong motivation to propose a sophisticated data representation that can capture the significant amount of details presented in a VHSR dataset and to search for a more powerful scheme suitable for multiple remotely sensed data segmentations. This thesis focuses on the development of a segmentation framework for multiple VHSR remotely sensed data. The emphases are on VHSR data model and segmentation strategy. Starting with the domain partition of a given remotely sensed dataset, a hierarchical data model characterizing the structures hidden in the dataset locally, regionally and globally is built by three random fields: Markova random field (MRF), strict stationary random field (RF) and label field. After defining prior probability distributions which should capture and characterize general and scene-specific knowledge about model parameters and the contextual structure of accurate segmentations, the Bayesian based segmentation framework, which can lead to algorithmic implementation for multiple remotely sensed data, is developed by integrating both the data model and the prior knowledge. To verify the applicability and effectiveness of the proposed segmentation framework, the segmentation algorithms for different types of remotely sensed data are designed within the proposed segmentation framework. The first application relates to SAR intensity image processing, including segmentation and dark spot detection by marked point process. In the second application, the algorithms for LiDAR point cloud segmentation and building detection are developed. Finally, texture and colour texture segmentation problems are tackled within the segmentation framework. All applications demonstrate that the proposed data model provides efficient representations for hierarchical structures hidden in remotely sensed data and the developed segmentation framework leads to successful data processing algorithms for multiple data and task such as segmentation and object detection.

The aim of this thesis is to develop a methodology for detection of naturally occurring offshore oil slicks originating from hydrocarbon seeps using satellite remote sensing techniques. In this scope, Synthetic Aperture Radar (SAR) imagery has been utilized. Case study area was Andrusov High in the Central Black Sea. Hydrocarbon seepage from tectonic or stratigraphic origin at the sea floor causes oily gas plumes to rise up to the sea surface. They form thin oil films on the sea surface called oil slicks. Presence of seeps and surface oil slicks for the offshore basins is a trace of depleted oil traps. Spatial distribution of oil slicks is closely related to sea waves, dominant wind patterns and weathering factors. Even though, there are oil slick detection techniques available with optical remote sensing, laser fluorosensors, and hyperspectral remote sensing, the most efficient results can be obtained from active microwave sensors like synthetic aperture radar (SAR). SAR sensors simply measure the backscattered radiation from the surface and show the roughness of the terrain. Oil slicks dampen the sea waves creating dark patches in the SAR image. In this context an adapted methodology has been proposed, including three levels namely / visual inspection, image filtering and object based fuzzy classification. With visual inspection, targets have been identified and subset scenes have been created. Subset scenes have been categorized into 3 cases based on contrast difference of dark spots to the surroundings. Then object based classification has been utilized with the fuzzy membership functions defined by extracted features of layer values, shape and texture from segmented and filtered SAR subsets. As a result, oil slicks have been discriminated from look-alikes which are the phenomena resembling oil slicks. The overall classification accuracy obtained by averaging three different cases is 83 % for oil slicks and 77 % for look-alikes. The results of this study can considered to be a preliminary work and supplementary information for determining the best operational procedure of offshore hydrocarbon exploration.

Soil moisture condition of a watershed plays a significant role in separation of rainfall into infiltration and surface runoff, and hence is a key parameter for the majority of physical hydrological models. Due to the large difference in dielectric constants of dry soil and water, microwave remote sensing and particularly the commonly available synthetic aperture radar is a potential tool for such studies. The main aim of this study is to produce the distributed soil moisture maps of a catchment from active microwave imagery. For this purpose, nine field trips are performed within a small basin in western Anatolia and point surface soil moisture values are collected with a Time Domain Reflectometer. The field studies are planned to match radar image acquisitions and accomplished over the water year of 2004 - 2005. In this context, first, the Dubois Model, a semi-empirical backscatter model is utilized in the reverse order to develop radar backscatter &amp / #8211 / soil roughness relationship and soil roughness maps of the study area are obtained. Then another relationship is built between radar backscatter and the three governing surface parameters: local incidence angle, soil moisture and soil roughness, which is later used in the soil moisture estimation methods. Depending on land use and vegetation cover condition, surface soil moisture maps of the catchment are produced by Backscatter Correction Factors, Water Cloud Model and Basin Indexes methods. In the last part of the study, the soil moisture maps of the basin are input to a semi-distributed hydrological model, HEC-HMS, as the initial soil moisture condition of a flood event simulation. In order to investigate the contribution of distributed initial soil moisture data on model outputs, simulation of the same flood event is also performed with the lumped initial soil moisture condition. Finally, a comparison between both the distributed and lumped model simulation outputs and with the observed data is carried out.

In the scope of this thesis, simulation-based analyses and comparative evaluation of Synthetic Aperture Radar (SAR) image formation techniques, namely Time Domain Correlation, Range Stacking, Range Doppler and Chirp Scaling algorithms, are presented. For this purpose, first, the fundamental concepts of SAR such as SAR geometry, resolution and signal properties are explained. A broadside SAR simulator that provides artificial raw data as an input to the algorithms is designed and implemented. Then, the mathematical background of the imaging algorithms discussed in the thesis is provided. Implementations of these algorithms and simulations are carried out using MATLAB&reg / . Finally, simulation results are presented and discussed to show the advantages and disadvantages of the algorithms.

Before launching into ultrasound research, it is important to recall that the ultimate goal is to provide the clinician with the best possible information needed to make an accurate diagnosis. Ultrasound images are inherently affected by speckle noise, which is due to image formation under coherent waves. Thus, it appears to be sensible to reduce speckle artifacts before performing image analysis, provided that image texture that might distinguish one tissue from another is preserved. The main goal of this thesis was the development of novel speckle suppression methods from medical ultrasound images in the multiscale wavelet domain. We started by showing, through extensive modeling, that the subband decompositions of ultrasound images have significantly non-Gaussian statistics that are best described by families of heavy-tailed distributions such as the alpha-stable. Then, we developed Bayesian estimators that exploit these statistics. We used the alpha-stable model to design both the minimum absolute error (MAE) and the maximum a posteriori (MAP) estimators for alpha-stable signal mixed in Gaussian noise. The resulting noise-removal processors perform non-linear operations on the data and we relate this non-linearity to the degree of non-Gaussianity of the data. We compared our techniques to classical speckle filters and current state-of-the-art soft and hard thresholding methods applied on actual ultrasound medical images and we quantified the achieved performance improvement. Finally, we have shown that our proposed processors can find application in other areas of interest as well, and we have chosen as an illustrative example the case of synthetic aperture radar (SAR) images. / Ο απώτερος σκοπός της έρευνας που παρουσιάζεται σε αυτή τη διδακτορική διατριβή είναι η διάθεση στην κοινότητα των κλινικών επιστημόνων μεθόδων οι οποίες να παρέχουν την καλύτερη δυνατή πληροφορία για να γίνει μια σωστή ιατρική διάγνωση. Οι εικόνες υπερήχων προσβάλλονται ενδογενώς από θόρυβο, ο οποίος οφείλεται στην διαδικασία δημιουργίας των εικόνων μέσω ακτινοβολίας που χρησιμοποιεί σύμφωνες κυματομορφές. Είναι σημαντικό πριν τη διαδικασία ανάλυσης της εικόνας να γίνεται απάλειψη του θορύβου με κατάλληλο τρόπο ώστε να διατηρείται η υφή της εικόνας, η οποία βοηθά στην διάκριση ενός ιστού από έναν άλλο. Κύριος στόχος της διατριβής αυτής υπήρξε η ανάπτυξη νέων μεθόδων καταστολής του θορύβου σε ιατρικές εικόνες υπερήχων στο πεδίο του μετασχηματισμού κυματιδίων. Αρχικά αποδείξαμε μέσω εκτενών πειραμάτων μοντελοποίησης, ότι τα δεδομένα που προκύπτουν από τον διαχωρισμό των εικόνων υπερήχων σε υποπεριοχές συχνοτήτων περιγράφονται επακριβώς από μη-γκαουσιανές κατανομές βαρέων ουρών, όπως είναι οι άλφα-ευσταθείς κατανομές. Κατόπιν, αναπτύξαμε Μπεϋζιανούς εκτιμητές που αξιοποιούν αυτή τη στατιστική περιγραφή. Πιο συγκεκριμένα, χρησιμοποιήσαμε το άλφα-ευσταθές μοντέλο για να σχεδιάσουμε εκτιμητές ελάχιστου απόλυτου λάθος και μέγιστης εκ των υστέρων πιθανότητας για άλφα-ευσταθή σήματα αναμεμειγμένα με μη-γκαουσιανό θόρυβο. Οι επεξεργαστές αφαίρεσης θορύβου που προέκυψαν επενεργούν κατά μη-γραμμικό τρόπο στα δεδομένα και συσχετίζουν με βέλτιστο τρόπο αυτή την μη-γραμμικότητα με τον βαθμό κατά τον οποίο τα δεδομένα είναι μη-γκαουσιανά. Συγκρίναμε τις τεχνικές μας με κλασσικά φίλτρα καθώς και σύγχρονες μεθόδους αυστηρού και μαλακού κατωφλίου εφαρμόζοντάς τες σε πραγματικές ιατρικές εικόνες υπερήχων και ποσοτικοποιήσαμε την απόδοση που επιτεύχθηκε. Τέλος, δείξαμε ότι οι προτεινόμενοι επεξεργαστές μπορούν να βρουν εφαρμογές και σε άλλες περιοχές ενδιαφέροντος και επιλέξαμε ως ενδεικτικό παράδειγμα την περίπτωση εικόνων ραντάρ συνθετικής διατομής.

This dissertation presents new methods for detecting dynamic and thermodynamic characteristics of Arctic sea ice using radar remote sensing. A new technique for sea ice motion detection from sequential satellite synthetic aperture radar (SAR) images was developed and thoroughly validated. The accuracy of the system is 0.43 km obtained from a comparison between SAR-derived ice motion vectors and in-situ sea ice beacon trajectories. For the first time, we evaluated ice motion tracking results derived from co-polarization (HH) and cross-polarization (HV) channels of RADARSAT-2 ScanSAR imagery and formulated a condition where the HV channel is more reliable than the HH channel for ice motion tracking. Sea ice motion is substantially controlled by surface winds. Two new models for ocean surface wind speed retrieval from C-band SAR data have been developed and validated based on a large body of statistics on buoy observations collocated and coincided with RADARSAT-1 and -2 ScanSAR images. The proposed models without wind direction input demonstrated a better accuracy than conventionally used algorithms. As a combination of the developed methods we designed a wind speed-ice motion product which can be a useful tool for studying sea ice dynamics processes in the marginal ice zone. To effectively asses the thermodynamic properties of sea ice advanced tools for modeling electromagnetic (EM) wave scattering from rough natural surfaces are required. In this dissertation we present a new analytical formulation for EM wave scattering from rough boundaries interfacing inhomogeneous media based on the first-order approximation of the small perturbation method. Available solutions in the literature represent special cases of our general solution. The developed scattering theory was applied to experimental data collected at three stations (with different snow thicknesses) in the Beaufort Sea from the research icebreaker Amundsen during the Circumpolar Flaw Lead system study. Good agreement between the model and experimental data were observed for all three case studies. Both model and experimental radar backscatter coefficients were considerably higher for thin snow cover (4 cm) compared to the thick snow cover case (16 cm). Our findings suggest that, winter snow thickness retrieval may be possible from radar observations under particular scattering conditions.

Pendant ces dix dernières années, les lancements successifs de satellites pour l´observation de la Terre dotés de capteurs SAR (Synthetic Aperture Radar) ont permis de montrer son fort potentiel, car ces systèmes sont capables de couvrir de vastes régions avec une résolution élevée, ce qui représente un avantage pour le suivi terrain. Cette thèse propose d’élargir les applications conventionnelles, en s’occupant d´investiguer le potentiel et les limites de l’interférométrie SAR pour la mesure de la déformation du terrain dans la région amazonienne, qui n’a pas encore été étudiée dans ce cadre spécifique. L´application consiste à estimer le déplacement de la surface de la Terre sur la zone urbaine de Manaus, la plus grande ville de l'État d'Amazonas au Brésil, qui représente un site important pour l'exploration pétrolière et gazière et pour le transport. Ce site est entouré par des écosystèmes fragiles, qui le rendent très sensible à la présence de l'industrie pétrolière. Dans ce contexte, une compréhension de la dynamique temporelle et de la distribution spatiale des phénomènes de néotectonique est fondamentale pour la définition de bonnes pratiques de gestion de l'environnement. Au niveau méthodologique, afin de lever les principales difficultés rencontrées pour l’application de l’interférométrie différentielle sur une pile de données Radarsat-1, une stratégie multi-échelles et “model free”, basée sur l´information de déformation non-linéaire au cours du temps est adoptée avec succès. La caractéristique essentielle de cette procédure est la séparation du signal de phase en différentes échelles spatiales pour simplifier la séparation des trois composantes de phase (topographie, atmosphère et déplacement). Cela conduit à une plus grande robustesse et permet l'inversion de composantes de phase pour les petites piles d’images. Au niveau géophysique, l’application de l’interférométrie à l’étude du déplacement du terrain est réalisée pour la première fois sur le milieu de l´Amazonie, en complétant les études de géologie structurale antérieures basées sur les mesures issues de la corrélation des images optiques et les mesures de terrain. En complément des connaissances antérieures, la présente étude apporte une information précise sur l'hypothèse de mouvements de la croûte récents liés aux activités néotectoniques du bassin de l'Amazonie. Les résultats indiquent une zone de mouvement de la croûte adjacente à une structure de drainage circulaire dans la ville de Manaus. Les images Radarsat-1 et 2 acquises sur cette région apportent une meilleure compréhension des activités géologiques et du mouvement de la croûte dans le bassin de l'Amazonie. / During the past decade, successive satellites launches for Earth observation with SAR (Synthetic Aperture Radar) sensors onboard have shown its potential, once these systems are able to cover large areas with high resolution, which represents an advantage for surface monitoring. The mass of data generated has enabled the development of radar interferometry and its relevance to the study of small deformations in urban areas or in fault zones, showing that the technique was able to focus on different spatial scales of deformation, as well as temporal scales ranging from a few weeks to more than a decade. Usually, these types of applications have been limited to non-equatorial regions of the world due to the presence of atmospheric disturbances that affect the radar signal. Given the large size and the remote location of tropical basins such as the Amazon, satellite observations remain as a viable approach to validate existing geophysical models. In this context, this thesis proposes to extend conventional applications, taking care to investigate the potential and limitations of SAR interferometry for the measurement of ground deformation in the Amazon region, not yet studied in this specific context. The research aim is to estimate the displacement at the surface of the Earth on the urban area of Manaus, the largest city in the state of Amazonas in Brazil, which is an important site for oil and gas exploration and the transport. This site is surrounded by fragile ecosystems, which make it very sensitive to the presence of the oil industry. Considering this, an understanding of the temporal dynamics and spatial distribution of neotectonic phenomena is fundamental to the development of best practice environmental management. At the methodological level, to overcome the major challenges of the application of differential interferometry on the data stack Radarsat-1, a multi-scale and "model free" strategy, based on the information of a non-linear deformation over time is passed successfully. The essential feature of this process is the separation of the phase signal into small and large scale spatial contributions to simplify the separation of the three phase components (topography, atmosphere and movement). This leads to a more robust processing and allows the phase component inversion for small piles of images. In the geophysical level, the application of interferometry to investigate the ground movement is performed for the first time in the middle of the Amazon, complementing previous studies of structural geology based on measurements from the correlation of optical images and field measurements. In addition to prior knowledge, this study provides accurate information on the hypothesis of recent crustal movements associated with neotectonic activities of the Amazon basin. The results indicate a range of motion of the adjacent crust structure circular drainage in the city of Manaus. The Radarsat-1 and 2 acquired in this region provide a better understanding of geological activity and crustal movement in the Amazon basin.