41 research outputs found
Assimilation d'images pour les fluides géophysiques
No full textUnderstanding and forecasting the evolution of geophysical fluids is a major scientific and societal challenge. A good forecast must take into account all available information on the studied system. These informations include models, observations and a priori knowledge. Data assimilation techniques combine all these informations in a consistent way to produce model inputs. During the last decades, many satellites were launched to increase the knowledge of earth. They produce, among others, image sequences showing the dynamical evolution of geophysical processes such as depressions and fronts. These images sequences are currently under-utilized in data assimilation. This thesis presents a consistent approach for taking into account image sequences in variational data assimilation. After a presentation of images, their current use and its limitation, we introduce the concepts of interpretation level, image space and image operator used for direct image sequences assimilation. We also propose a new approach of regularization based on generalized diffusion for ill-posed inverse problems. Preliminary results on image processing and image sequences assimilation show a promising approach that solve most of the problems encountered with classical approaches of regularization.La compréhension et la prévision de l'évolution des fluides géophysiques sont d'une importance capitale et constituent un domaine de recherche scientifique aux enjeux conséquents. Une bonne prévision est basée sur la prise en compte de toutes les informations disponibles sur le système considéré. Ces informations incluent les modèles, les observations et les connaissances a priori. L'assimilation de données permet de les combiner de façon optimale pour déterminer les entrées du modèle. Les dernières décennies ont vu croître en densité et en qualité la couverture satellitaire produisant, entre autres, des séquences d'images montrant l'évolution dynamique de certains phénomènes géophysiques tels que les dépressions et les fronts. Ces séquences d'images sont jusqu'à présent sous-utilisées en assimilation de données. Cette thèse propose une extension de l'assimilation variationnelle de données aux observations de type séquence d'images. Après avoir présenté les images, leur utilisation actuelle et ses limites, nous introduisons les notions de niveau d'interprétation, d'espaces et d'opérateur image. Ces notions sont utilisées pour formuler l'assimilation directe de séquences d'images. Nous proposons également une nouvelle approche de régularisation par diffusion généralisée pour les problèmes inverses. Les résultats préliminaires en traitement d'images et en assimilation directe de séquence d'images montrent une méthode prometteuse qui résout la plupart des problèmes rencontrés avec les approches classiques de régularisation
Image Assimilation for Geophysical Fluids: the ADDISA project. Active contours and Vortex Motion
No full textInternational audienc
Régularisation par diffusion généralisée pour les problèmes inverses mal posés : application au traitement d'images en géophysique
No full textNational audienceL'estimation du mouvement par flot optique est sujet au problème d'ouverture : seule la composante de la vitesse normale aux iso-contours peut être estimée. Pour faire face à ce problème, on utilise des techniques de régularisation. De façon usuelle, on complète la fonction coût par un terme de pénalisation de l'écart à la régularité. L'utilisation de cette approche pose trois principaux problèmes : le choix du paramètre de pondération du terme de pénalité - la convergence des algorithmes de minimisation en présence d'une fonction coût composite - l'interprétation physique de la régularisation. Dans ce papier, nous proposons une nouvelle approche pour la régularisation des problèmes inverses mal posés; toute l'information de régularisation est utilisée pour définir une norme appropriée à la fonction coût par l'intermédiaire d'une fonction de confiance. On se débarrasse ainsi des termes supplémentaires de régularisation qui rendent la fonction coût composite. L'information de régularité et la fonction de confiance permettent de définir un préconditionnement approprié qui accélère la convergence tout en se passant des paramètres de pondération lié à la pénalisation. La fonction de confiance introduite ici permet de donner une interprétation physique valable aux termes de régularisation. Nous proposons une dérivation simplifiée de la fonction de confiance dans le cas de l'estimation du mouvement par flot optique en prenant en compte le problème d'ouverture. Les résultats obtenus par cette nouvelle approche sont présentés dans le cadre de l'estimation du mouvement ainsi qu'une comparaison avec les approches usuelles. Ces résultats montrent la supériorité de la nouvelle approche qui semble prometteuse pour la régularisation des problèmes inverses mal posés
Direct Assimilation of Image Sequences
No full textPosterNational audienceIn order to forecast the evolution of a dynamical system such as geophysical fluids ocean, atmosphere, continental waters), all the available information have to be accounted for. They are of very different nature: set of non linear PDE (mathematical-type information), in situ measurements and remote sensing (physical-type information), statistical and qualitative informations. The forecast is produced through a model integration starting from an initial state, from which the system evolution is very sensitive. Consequently, the issue is to evaluate the initial state in a consistent manner from all this heterogeneous sources of information. At the beginning of the 80s, techniques coming from the optimal control theory were proposed to achieve this task. These techniques are now adopted by the main numerical weather forecast centers. For few decades, a large number satellites dedicated to earth observation has been launched, in order the improve our knowledge of the atmosphere and the oceans. They provide, among other things, numerous sequences of images. These sequences clearly have a strong predictive potential due to the fact that they contain information about the dynamics of the observed system. Currently, this kind of information is unfortunately not used in an optimal manner in conjunction with the numerical models. This poster presents an extension of the optimal control based techniques to the assimilation of images. A quadratic term measuring the misfit between the images equivalent produced by the model and the observed images is introduced in the usual cost function
Assimilation of Image Sequences in Numerical Models
No full textUnderstanding and forecasting the evolution of geophysical fluids is a major scientific and societal challenge. Forecasting algorithms should take into account all the available informations on the considered dynamical system. The Variational Data Assimilation (VDA) technique combines in a consistent way all these informations in an Optimality System in order to reconstruct the model inputs. VDA is currently used by the major meteorological centres. During the last two decades about thirty satellites were launched to improve the knowledge of the atmosphere and of the oceans. They continuously provide a huge amount of data that are still underused by numerical forecast systems. In particular, the dynamical evolution of some meteorological or oceanic features (such as eddies, fronts, \dots) that a human vision may easily detect is not optimally taken into account in realistic applications of VDA. Image Assimilation in VDA framework can be performed using \textit{pseudo-observation} techniques : they provide some apparent velocity fields which are assimilated as classical observations. These measurements are obtained by some external procedures which are decoupled with the considered dynamical system. In this paper, we suggest a more consistent approach which directly incorporates image sequences into the Optimality System
A variational data assimilation system for the range dependent acoustic model using the representer method: Theoretical derivations
No full textA new approach for regularization of inverse problems in images processing
No full textInternational audienceOptical flow motion estimation from two images is limited by the aperture problem. A method to deal with this problem is to use regularization techniques. Usually, one adds a regularization term with appriopriate weighting parameter to the optical flow cost funtion. Here, we suggest a new approach to regularization for optical flow motion estimation. In this approach, all the regularization informations are used in the definition of an appropriate norm for the cost function via a trust function to be defined, one does not ever need weighting parameter. A simple derivation of such a trust function from images is proposed and a comparison with usual approaches is presented. These results show the superiority of such approach over usual ones
Assimilation of Image Sequences in Numerical Models
No full textUnderstanding and forecasting the evolution of geophysical fluids is a major scientific and societal challenge. Forecasting algorithms should take into account all the available informations on the considered dynamical system. The Variational Data Assimilation (VDA) technique combines in a consistent way all these informations in an Optimality System in order to reconstruct the model inputs. VDA is currently used by the major meteorological centres. During the last two decades about thirty satellites were launched to improve the knowledge of the atmosphere and of the oceans. They continuously provide a huge amount of data that are still underused by numerical forecast systems. In particular, the dynamical evolution of some meteorological or oceanic features (such as eddies, fronts, \dots) that a human vision may easily detect is not optimally taken into account in realistic applications of VDA. Image Assimilation in VDA framework can be performed using \textit{pseudo-observation} techniques : they provide some apparent velocity fields which are assimilated as classical observations. These measurements are obtained by some external procedures which are decoupled with the considered dynamical system. In this paper, we suggest a more consistent approach which directly incorporates image sequences into the Optimality System
Vector field regularization by generalized diffusion
No full textRegularization is a common procedure when dealing with inverse problems. Because of the ill-posedness of many inverse problems, one needs to add some constraints as regularization to the problem in order to get a satisfactory solution. A difficulty when using multiple constraints is to properly choose a weighting parameter for each constraint. We propose here a vector field regularization method that combines in a single constraint the two well-known regularization methods namely Tikhonov regularization and smoothing regularization. The particularity of this new method is that one have only one balance parameter to determine. We also suggest a robust implementation of the proposed method based on the equivalent generalized diffusion equation in some particular cases. This implementation is illustrated on a set of vector fields of fluid motio
Assimilation of Image Sequences in Numerical Models
No full textInternational audienceUnderstanding and forecasting the evolution of geophysical fluids is a major scientific and societal challenge. Forecasting algorithms should take into account all the available information on the considered dynamic system. The variational data assimilation (VDA) technique combines all these informations in an optimality system (O.S.) in a consistent way to reconstruct the model inputs. VDA is currently used by the major meteorological centres. During the last two decades about 30 satellites were launched to improve the knowledge of the atmosphere and of the oceans. They continuously provide a huge amount of data that are still underused by numerical forecast systems. In particular, the dynamic evolution of certain meteorological or oceanic features (such as eddies, fronts, etc.) that the human vision may easily detect is not optimally taken into account in realistic applications of VDA. Image Assimilation in VDA framework can be performed using 'pseudo-observation' techniques: they provide apparent velocity fields, which are assimilated as classical observations. These measurements are obtained by certain external procedures, which are decoupled with the considered dynamic system. In this paper, we suggest a more consistent approach, which directly incorporates image sequences into the O.S
