La résonance paramagnétique électronique est une technologie permettant de localiser et de caractériser les radicaux libres, fondée sur la propriété de résonance des électrons libres lorsqu’ils sont placés dans un champ magnétique spécifique. Afin d’augmenter la qualité des reconstructions obtenues par des dispositifs d’imagerie de résonance paramagnétique électronique, ce travail propose l’utilisation de méthodes variationnelles pour inverser le modèle de formation des images, qui combine une convolution avec une transformée de Radon. La fonctionnelle proposée repose sur la norme L2 pour le terme d’attache aux données, et sur la variation totale et une seminorme de Besov pour le terme de régularisation. La seminorme de Besov est implémentée grâce à la transformée en curvelets et à la norme L1 qui permet d’appliquer un critère de parcimonie. Les propriétés de ces termes de régularisation permettent de reconstruire des images à la fois pertinentes dans les zones où l’acquisition des données est insuffisante, notamment sur les bords, et suffisamment détaillées dans les zones où l’échantillon est texturé. L’augmentation de la qualité des images reconstruites permet d’envisager des acquisitions sur des durées réduites, ouvrant la voie à des expériences in vivo ou cliniques actuellement limitées par des durées d’acquisition de l’ordre de plusieurs dizaines de minutes. Les algorithmes de minimisation primal-dual de Chambolle-Pock et FISTA sont utilisés pour résoudre les problèmes d’optimisation que pose l’utilisation de méthodes variationnelles. L’étude détaillée du modèle direct permet de mettre en évidence une structure de Toeplitz, dont les propriétés sont utilisées pour résoudre le problème inverse en évitant le recours à la rétroprojection filtrée ou aux transformées de Fourier non-uniformes. Des simulations numériques sont menées sur le fantôme de Shepp-Logan, et valident le modèle proposé qui surpasse à la fois visuellement et quantitativement les techniques de reconstruction couramment utilisées, combinant déconvolution et rétroprojection filtrée. Des reconstructions menées sur des acquisitions réelles, consistant en un échantillon papier d’une encre paramagnétique et en une phalange distale irradiée, valident par l’expérience le choix des fonctionnelles utilisées pour inverser le modèle direct. La grande souplesse de la méthode variationnelle proposée permet d’adapter la fonctionnelle au problème de la séparation de sources qui se pose lorsque deux molécules paramagnétiques différentes sont présentes au sein d’un même échantillon. La fonctionnelle proposée permet de séparer les deux molécules dans le cadre d’une acquisition classique d’imagerie de résonance paramagnétique électronique, ce qui n’était possible jusqu’alors que sur des acquisitions dites hyperspectrales beaucoup plus gourmandes en temps. / Spatial electron paramagnetic resonance imaging (EPRI) is a recent method to localize and characterize free radicals in vivo or in vitro, leading to applications in material and biomedical sciences. To improve the quality of the reconstruction obtained by EPRI, a variational method is proposed to inverse the image formation model. It is based on a least-square data-fidelity term and the total variation and Besov seminorm for the regularization term. To fully comprehend the Besov seminorm, an implementation using the curvelet transform and the L1 norm enforcing the sparsity is proposed. It allows our model to reconstruct both image where acquisition information are missing and image with details in textured areas, thus opening possibilities to reduce acquisition times. To implement the minimization problem using the algorithm developed by Chambolle and Pock, a thorough analysis of the direct model is undertaken and the latter is inverted while avoiding the use of filtered backprojection (FBP) and of non-uniform Fourier transform. Numerical experiments are carried out on simulated data, where the proposed model outperforms both visually and quantitatively the classical model using deconvolution and FBP. Improved reconstructions on real data, acquired on an irradiated distal phalanx, were successfully obtained. Due to its great versatility, the variational approach is easily extended to the source separation problem which happens when two different paramagnetic species are present in the sample. The objective function is consequently modified, and a classic EPRI acquisition yields two images, one for each species. Until now, source separation could only be applied to hyperspectral EPRI data, much more costly in acquisition time.

Les mesures de vibrations sont l’une des données les plus pertinentes pour détecter des anomalies sur les moteurs. Les vibrations sont acquises sur banc d’essai en phase d’accélération et de décélération pour assurer la fiabilité du moteur à la sortie de la chaine de production. Ces données temporelles sont converties en spectrogrammes pour permettre aux experts d’effectuer une analyse visuelle de ces données et de détecter les différentes signatures atypiques. Les sources vibratoires correspondent à des raies sur les spectrogrammes. Dans cette thèse, nous avons mis en place un outil d’aide à la décision automatique pour analyser les spectrogrammes et détecter tout type de signatures atypiques, ces signatures ne proviennent pas nécessairement d’un endommagement du moteur. En premier lieu, nous avons construit une base de données numérique de spectrogrammes annotés. Il est important de noter que les signatures inusuelles sont variables en forme, intensité et position et se trouvent dans un faible nombre de données. Par conséquent, pour détecter ces signatures, nous caractérisons les comportements normaux des spectrogrammes, de manière analogue aux méthodes de détection de nouveautés, en représentant les patchs des spectrogrammes sur des dictionnaires comme les curvelets et la Non-negative matrix factorization (NMF), ainsi qu’en estimant la distribution de chaque point du spectrogramme à partir de données normales dépendamment ou non de leur voisinage. La détection des points atypiques est réalisée par comparaison des données tests au modèle de normalité estimé sur des données d’apprentissage normales. La détection des points atypiques permet la détection des signatures inusuelles composées par ces points. / Vibration measurements are one of the most relevant data for detecting anomalies in engines. Vibrations are recorded on a test bench during acceleration and deceleration phases to ensure the reliability of every flight engine at the end of the production line. These temporal signals are converted into spectrograms for experts to perform visual analysis of these data and detect any unusual signature. Vibratory signatures correspond to lines on the spectrograms. In this thesis, we have developed a decision support system to automatically analyze these spectrograms and detect any type of unusual signatures, these signatures are not necessarily originated from a damage in the engine. Firstly, we have built a numerical spectrograms database with annotated zones, it is important to note that data containing these unusual signatures are sparse and that these signatures are quite variable in shape, intensity and position. Consequently, to detect them, like in the novelty detection process, we characterize the normal behavior of the spectrograms by representing patches of the spectrograms in dictionaries such as the curvelets and the Non-negative matrix factorization (NMF) and by estimating the distribution of every points of the spectrograms with normal data depending or not of the neighborhood. The detection of the unusual points is performed by comparing test data to the model of normality estimated on learning normal data. The detection of the unusual points allows the detection of the unusual signatures composed by these points.

Two complementary solution strategies to the least-squares migration problem with sparseness- & continuity constraints are proposed. The applied formalism explores the sparseness of curvelets on the reflectivity and their invariance under the demigration migration operator. Sparseness is enhanced by (approximately) minimizing a (weighted) l1-norm on the curvelet coefficients. Continuity along imaged reflectors is brought out by minimizing the anisotropic difussion or total variation norm which penalizes variations along and in between reflectors. A brief sketch of the theory is provided as well as a number of synthetic examples. Technical details on the implementation of the optimization strategies are deferred to an accompanying paper: implementation.

We present an alternative method of extracting production related differences from time-lapse seismic data sets. Our method is not based on the actual subtraction of the two data sets, risking the enhancement of noise and introduction of artifacts due to local phase rotation and slightly misaligned events. Rather, it mutes events of the monitor survey with respect to the baseline survey based on the magnitudes of coefficients in a sparse and local atomic decomposition. Our technique is demonstrated to be an effective tool for enhancing the time-lapse signal from surveys which have been cross-equalized

We propose a method for seismic data interpolation based on 1) the reformulation of the problem as a stable signal recovery problem and 2) the fact that seismic data is sparsely represented by curvelets. This method does not require information on the seismic velocities. Most importantly, this formulation potentially leads to an explicit recovery condition. We also propose a large-scale problem solver for the l1-regularization minimization involved in the recovery and successfully illustrate the performance of our algorithm on 2D synthetic and real examples.

In this overview of the recent Curvelet Reconstruction with Sparsity-promoting Inversion (CRSI) method, we present our latest 2-D and 3-D interpolation results on both synthetic and real datasets. We compare these results to interpolated data using other existing methods. Finally, we discuss the challenges related to sparsity-promoting solvers for the large-scale problems the industry faces.

In this abstract, we present a novel primary-multiple separation scheme which makes use of the sparsity of both primaries and multiples in a transform domain, such as the curvelet transform, to provide estimates of each. The proposed algorithm utilizes seismic data as well as the output of a preliminary step that provides (possibly) erroneous predictions of the multiples. The algorithm separates the signal components, i.e., the primaries and multiples, by solving an optimization problem that assumes noisy input data and can be derived from a Bayesian perspective. More precisely, the optimization problem can be arrived at via an assumption of a weighted Laplacian distribution for the primary and multiple coefficients in the transform domain and of white Gaussian noise contaminating both the seismic data and the preliminary prediction of the multiples, which both serve as input to the algorithm.

In this paper, we recover the amplitude of a seismic image by approximating the normal (demigrationmigration)operator. In this approximation, we make use of the property that curvelets remain invariant under the action of the normal operator. We propose a seismic amplitude recovery method that employs an eigenvalue like decomposition for the normal operator using curvelets as eigen-vectors. Subsequently, we propose an approximate non-linear singularity-preserving solution to the least-squares seismic imaging problem with sparseness in the curvelet domain and spatial continuity constraints. Our method is tested with a reverse-time ’wave-equation’ migration code simulating the acoustic wave equation on the SEG-AA salt model.

In this abstract, we present a nonlinear curvelet-based sparsitypromoting formulation for the primary-multiple separation problem. We show that these coherent signal components can be separated robustly by explicitly exploting the locality of curvelets in phase space (space-spatial frequency plane) and their ability to compress data volumes that contain wavefronts. This work is an extension of earlier results and the presented algorithms are shown to be stable under noise and moderately erroneous multiple predictions.

An \emph {explicit} algorithm for the extrapolation of one-way wavefields is proposed which combines recent developments in information theory and theoretical signal processing with the physics of wave propagation. Because of excessive memory requirements, explicit formulations for wave propagation have proven to be a challenge in {3-D}. By using ideas from ``\emph{compressed sensing}'', we are able to formulate the (inverse) wavefield extrapolation problem on small subsets of the data volume{,} thereby reducing the size of the operators. According {to} compressed sensing theory, signals can successfully be recovered from an imcomplete set of measurements when the measurement basis is \emph{incoherent} with the representation in which the wavefield is sparse. In this new approach, the eigenfunctions of the Helmholtz operator are recognized as a basis that is incoherent with curvelets that are known to compress seismic wavefields. By casting the wavefield extrapolation problem in this framework, wavefields can successfully be extrapolated in the modal domain via a computationally cheaper operatoion. A proof of principle for the ``compressed sensing'' method is given for wavefield extrapolation in {2-D}. The results show that our method is stable and produces identical results compared to the direct application of the full extrapolation operator.