This Ph.D. thesis aims at evaluating the interest of using multivariate distributions for the analysis of heterogeneous images. The considered heterogeneous data are composed of images acquired by different sensors (including optical, radar and hyperspectral sensors) and possibly of an object database (containing roads, building, etc.). The applications considered in this thesis are mainly image registration, change detection and database updating. All these applications require to define a similarity measure between the different images or between features estimated from these images (called modalities).

Remote sensing images are images of the Earth surface acquired from satellites or airborne equipment. These images are beco-ming widely available nowadays, with many commercial and non-commercial services pro-viding them. The technology of the sensors required to capture this kind of images is evol-ving fast. Not only classical sensors are impro-ving in terms of resolution and noise level, but also new kinds of sensors are proving to be useful. Multispectral image sensors are stan-dard nowadays, synthetic aperture radar (SAR) images are very popular, and hyperspectral sen-sors are receiving more and more attention in many applications. One of the main applications of remote sensing images is the detection of changes in multitem-poral datasets, i.e., detecting changes in images of the same area acquired at different times. Change detection for images acquired by ho-mogeneous sensors has been of interest for a long time. However the wide range of different sensors found in remote sensing makes the de-tection of changes in images acquired by hete-rogeneous sensors an interesting challenge. The main interest of this thesis is to study statistical approaches to detect changes in images acquired by heterogeneous sensors.

L'utilisation des méthodes d'accès aléatoires basées sur le codage réseau couche physique ainsi que la suppression successive des interférences s'avère intéressante dans le monde des réseaux satellitaires. Des techniques telles que CRDSA et CSA proposées par l'ESA, ainsi que MuSCA proposée par l'ISAE permettent l'augmentation du débit et la réduction des délais sur un lien retour, par rapport aux méthodes d'accès aléatoires traditionnelles. Notre travail consiste à étudier les effets du canal de transmission sur ces méthodes et de proposer une amélioration de CRDSA au niveau du récepteur, appelée MARSALA. Les performances globales du système sont analysées en termes de taux d'erreur paquets et de débit obtenu à l'aide d'implémentations sur le simulateur de réseaux par satellite à Thales Alenia Space.

Nous commençons par une courte introduction des principes du positionnement GNSS précis par mesures de phase dites RTK (Real-Time Kinematic) et PPP (Precise Point Positioning). Ensuite nous présentons les résultats actuels du projet COPNAV en cours de l'action PTP de TESA. Le signal des satellites GNSS est poursuivi par des boucles à verrouillage de code, fréquence et phase. Les sorties utilisées pour le positionnement conventionnel sont celles fournies par la boucle de code. Pour le positionnement précis, on utilise les mesures de phase, moins bruitées et en particulier moins sensibles aux multitrajets. Ces mesures sont cependant difficiles d’exploitation à cause du nombre entier de cycle qu'il faut estimer (résolution d’ambiguïté), la phase étant mesurée à 2 près. Les méthodes de positionnement précis GNSS telles que RTK ou PPP se distinguent par les moyens et informations supplémentaires nécessaires à l’estimation des ambiguïtés et à la correction des mesures pour arriver à une précision centimétrique. Mais ces performances sont habituellement obtenues avec des récepteurs de haute qualité, dans des environnements dégagés. En environnements contraints (urbain, intérieur, etc.), la disponibilité du positionnement précis baisse considérablement du fait des fréquentes pertes de signal provoquant des sauts de cycle. Nous présenterons les principes de la technologie RTK et les résultats que nous avons obtenus en utilisant le logiciel RTKLIB avec un récepteur bas coût U-Blox NEO 7, dans des environnements contraints (urbain et urbain dense), associée à une méthode innovante de compensation des sauts de cycle.

Satellite telecommunications have seen a tremendous increase in interest, due to its ability to reduce the digital divide. In fact, a geostationary satellite can take advantage of its very wide coverage and high capacity to reach areas where deployment of a terrestrial network would not be possible, such as transports, or too expensive to be profitable, as in remote areas. Traditionally focused on digital television broadcasting, the latest generation of standards have evolved to reflect those new needs, dealing extensively with the transmission of interactive data, particularly by natively supporting Internet protocols. Scheduling has arisen as a major issue of those modern systems, since it has to deal with to highly uncorrelated processes : demand and capacity. Demand, on one side, evolves with user's needs, and therefore with the applications they are using : video, voice or data. Capacity, on the other side, depends on meteorological conditions over the satellite's cover, as the frequencies used in such systems are very sensitive to wet atmosphere attenuation. This thesis aims to study the problem of scheduling and resource allocation, hoping to achieve a service that can match with terrestrial networks in terms of services, while showing the best possible performances. If numerous solutions were proposed on this topic, none is taking into account all of the current system's constraints. In addition to the variable nature of system's capacity, the conjunction of variable demand and quality of service constraints constitutes an additional issue. Furthermore, we have to consider the practicability of our solution in a real-time context, necessary if we aim for industrial use. We have first developed a scheduler architecture for the Forward link, based on utility functions, thus allowing a simple formulation of the capacity versus demand compromise. We show, through a detailed low-complexity implementation and accurate simulations, how our algorithm could be used eciently in an industrial context. We then focus on the Return link, where we propose a resource allocation method, taking into account quality of service and quality of transmission jointly to deliver an ecient yet consistent resource allocation. Simulations show that our algorithm achieves a better efficiency and traffic handling than reference solutions presented in the literature.

Satellite telecommunications have seen a tremendous increase in interest, due to its ability to reduce the digital divide. In fact, a geostationary satellite can take advantage of its very wide coverage and high capacity to reach areas where deployment of a terrestrial network would not be possible, such as transports, or too expensive to be profitable, as in remote areas. Traditionally focused on digital television broadcasting, the latest generation of standards have evolved to reflect those new needs, dealing extensively with the transmission of interactive data, particularly by natively supporting Internet protocols. Scheduling has arisen as a major issue of those modern systems, since it has to deal with to highly uncorrelated processes : demand and capacity. Demand, on one side, evolves with user's needs, and therefore with the applications they are using : video, voice or data. Capacity, on the other side, depends on meteorological conditions over the satellite's cover, as the frequencies used in such systems are very sensitive to wet atmosphere attenuation. This thesis aims to study the problem of scheduling and resource allocation, hoping to achieve a service that can match with terrestrial networks in terms of services, while showing the best possible performances. If numerous solutions were proposed on this topic, none is taking into account all of the current system's constraints. In addition to the variable nature of system's capacity, the conjunction of variable demand and quality of service constraints constitutes an additional issue. Furthermore, we have to consider the practicability of our solution in a real-time context, necessary if we aim for industrial use. We have first developed a scheduler architecture for the Forward link, based on utility functions, thus allowing a simple formulation of the capacity versus demand compromise. We show, through a detailed low-complexity implementation and accurate simulations, how our algorithm could be used efficiently in an industrial context. We then focus on the Return link, where we propose a resource allocation method, taking into account quality of service and quality of transmission jointly to deliver an efficient yet consistent resource allocation. Simulations show that our algorithm achieves a better efficiency and traffic handling than reference solutions presented in the literature.

Procédé de génération d'un banc de filtre (301,302,...30K) pour la réception d'un signal modulé par une modulation à phase continue, ledit signal modulé pouvant être décomposé comme une somme d'une pluralité de signaux modulés en amplitude, chaque signal modulé en amplitude étant exprimé par un produit entre un pseudo-symbole complexe et une composante temporelle de forme d'onde prédéfinie en fonction des paramètres de la modulation, ledit procédé comprenant les étapes suivantes : - Evaluer, sur une durée T égale à la durée d'un symbole, toutes les formes d'ondes possibles du signal à partir des paramètres de la modulation à phase continue et de la décomposition sous la forme d'une somme d'une pluralité de signaux modulés en amplitude, - Retenir toutes les formes d'ondes évaluées qui sont différentes entre elles, - Construire un banc de filtre (301,302,...30K) composé d'une pluralité de filtres dont les réponses temporelles sont égales aux formes d'ondes retenues, limitées à une durée égale à la durée d'un symbole.

The well-known conventional Weighted Least Squares (WLS) and extended Kalman filter (EKF) are the standard estimation methods for positioning with GNSS measurements. However, these estimators are not optimal when the GNSS measurements become contaminated by non-Gaussian errors including multipath (MP) and nonline-of-sight (NLOS) biases. In this paper, we use additional information of the geometric environment provided by a 3D model to build-up a robust solution against biases which may be summed up from MP and NLOS signals in urban environments. We first use a 3D city model to predict lower and upper bounds of these biases. Then, we integrate this information in the position estimation problem. We investigated in two ways of making use of this additional information: the first one is to consider these biases as additive noise and exploiting the bounds to end up with a constrained state estimation by WLS or Kalman filter. The second way is to investigate in the maximum likelihood estimation of both the MP-NLOS bias and the state ending up with a less accurate but acceptable solution. Test results using real GPS signal in Toulouse show that these estimators capable of improving the positioning accuracy compared to the conventional WLS if the NLOS bounds are well-chosen.

The dictionary learning problem aims at finding a dictionary of atoms that best represents an image according to a given objective. The most usual objective consists of representing an image or a class of images sparsely. Most algorithms performing dictionary learning iteratively estimate the dictionary and a sparse representation of images using this dictionary. Dictionary learning has led to many state of the art algorithms in image processing. However, its numerical complexity restricts its use to atoms with a small support since the computations using the constructed dictionaries require too much resources to be deployed for large scale applications. In order to alleviate these issues, this paper introduces a new strategy to learn dictionaries composed of atoms obtained as a composition of K convolutions with S-parse kernels. The dictionary update step associated with this strategy is a non-convex optimization problem. We reformulate the problem in order to reduce the number of its irrelevant stationary points and introduce a Gauss-Seidel type algorithm, referred to as Alternative Least Square Algorithm, for its resolution. The search space of the considered optimization problem is of dimension KS, which is typically smaller than the size of the target atom and is much smaller than the size of the image. The complexity of the algorithm is linear with regard to the size of the image. Our experiments show that we are able to approximate with a very high accuracy many atoms such as modified DCT, curvelets, sinc functions or cosines when K is large (say K = 10). We also argue empirically that, maybe surprisingly, the algorithm generally converges to a global minimum for large values of K and S.

Increasing the sampling rate of Analog-to-Digital Converters (ADC) is a main challenge in many fields and especially in telecommunications. Time-Interleaved ADCs (TI-ADC) were introduced as a technical solution to reach high sampling rates by time-interleaving several low-rate ADCs at the price of a perfect synchronization. Indeed, as the inverse operation of DAC is derived under the assumption of uniform sampling, the desynchronization must be compensated upstream with expensive hardware corrections of the sampling device. In this paper, we propose an alternative framework for TI-ADCs based on a more flexible sampling scheme, known as Periodic Non-uniform Sampling (PNS), that takes into account the desynchronization at the reconstruction stage when the induced delay is known, thus avoiding hardware corrections. The main contribution of this paper is to propose two different strategies for desynchronization estimation, one based on auto-calibration, the other on blind estimation. The proposed method performance is evaluated in terms of signal reconstruction error.