The sparse coding is approximation/representation of signals with the minimum number of coefficients using an overcomplete set of elementary functions. This kind of approximations/ representations has found numerous applications in source separation, denoising, coding and compressed sensing. The adaptation of the sparse approximation framework to the coding problem of signals is investigated in this thesis. Open problems are the selection of appropriate models and their orders, coefficient quantization and sparse approximation method. Some of these questions are addressed in this thesis and novel methods developed. Because almost all recent communication and storage systems are digital, an easy method to compute quantized sparse approximations is introduced in the first part. The model selection problem is investigated next. The linear model can be adapted to better fit a given signal class. It can also be designed based on some a priori information about the model. Two novel dictionary selection methods are separately presented in the second part of the thesis. The proposed model adaption algorithm, called Dictionary Learning with the Majorization Method (DLMM), is much more general than current methods. This generality allowes it to be used with different constraints on the model. Particularly, two important cases have been considered in this thesis for the first time, Parsimonious Dictionary Learning (PDL) and Compressible Dictionary Learning (CDL). When the generative model order is not given, PDL not only adapts the dictionary to the given class of signals, but also reduces the model order redundancies. When a fast dictionary is needed, the CDL framework helps us to find a dictionary which is adapted to the given signal class without increasing the computation cost so much. Sometimes a priori information about the linear generative model is given in format of a parametric function. Parametric Dictionary Design (PDD) generates a suitable dictionary for sparse coding using the parametric function. Basically PDD finds a parametric dictionary with a minimal dictionary coherence, which has been shown to be suitable for sparse approximation and exact sparse recovery. Theoretical analyzes are accompanied by experiments to validate the analyzes. This research was primarily used for audio applications, as audio can be shown to have sparse structures. Therefore, most of the experiments are done using audio signals.

Two inverse problems are investigated in this dissertation, taking into account both the spatial and temporal aspects. The first problem addresses the under determined image reconstruction problem for dynamic SPECT. The quality of the reconstructed image is often limited due to having fewer observations than the number of voxels. The proposed algorithms make use of the generalized α-divergence function to improve the estimation performance. The first algorithm is based on an alternating minimization framework to minimize a regularized α-divergence objective function. We demonstrate that selecting an adaptive α policy depending on the time evolution of the voxels gives better performance than a fixed α assignment. The second algorithm is based on Newton's method. A regularized approach has been taken to avoid stability issues. Newton's method is generally computationally demanding due to the complexity associated with inverting the Hessian matrix. A fast Newton-based method is proposed using majorization-minimization techniques that diagonalize the Hessian matrix. In dynamically evolving systems, the prediction matrix plays an important role in the estimation process. An estimation technique is proposed to estimate the prediction matrix using the α-divergence function. The simulation results show that our algorithms provide better performance than the techniques based on the Kullback-Leibler distance. The second problem is the recovery of data transmitted over free-space optical communication channels using orbital angular momentum (OAM). In the presence of atmospheric turbulence, crosstalk occurs among OAM optical modes resulting in an error floor at a relatively high bit error rate. The modulation format considered for the underlying problem is Q-ary pulse position modulation (PPM). We propose and evaluate three joint detection strategies to overcome the OAM crosstalk problem: i) maximum likelihood sequence estimation (MLSE). ii) Q-PPM factor graph detection. iii) branch-and-bound detection. We compare the complexity and the bit-error-rate performance of these strategies in realistic scenarios.

In contrast with traditional extended depth-of-field approaches, we propose a depth-based deconvolution technique that realizes the depth-variant nature of the point spread function of an ordinary fixed-focus camera. The developed technique brings a single blurred image to focus at different depth planes which can be stitched together based on a depth map to output a full-focus image. Strategies to suppress the deconvolution's ringing artifacts are implemented on three levels: block tiling to eliminate boundary artifacts, reference maps to reduce ringing initiated by sharp edges, and depth-based masking to mitigate artifacts raised by neighboring depth-transition surfaces. The performance is validated numerically for planar and multidepth objects. (C) 2016 Society of Photo-Optical Instrumentation Engineers (SPIE)

This work employs statistical and Bayesian techniques to analyze mathematical forward models with several sources of uncertainty. The forward models usually arise from phenomenological and physical phenomena and are expressed through regression-based models or partial differential equations (PDEs) associated with uncertain parameters and input data. One of the critical challenges in real-world applications is to quantify uncertainties of the unknown parameters using observations. To this purpose, methods based on the likelihood function, and Bayesian techniques constitute the two main statistical inferential approaches considered here. Two problems are studied in this thesis. The first problem is the prediction of fatigue life of metallic specimens. The second part is related to inverse problems in linear PDEs. Both problems require the inference of unknown parameters given certain measurements. We first estimate the parameters by means of the maximum likelihood approach. Next, we seek a more comprehensive Bayesian inference using analytical asymptotic approximations or computational techniques. In the fatigue life prediction, there are several plausible probabilistic stress-lifetime (S-N) models. These models are calibrated given uniaxial fatigue experiments. To generate accurate fatigue life predictions, competing S-N models are ranked according to several classical information-based measures. A different set of predictive information criteria is then used to compare the candidate Bayesian models. Moreover, we propose a spatial stochastic model to generalize S-N models to fatigue crack initiation in general geometries. The model is based on a spatial Poisson process with an intensity function that combines the S-N curves with an averaged effective stress that is computed from the solution of the linear elasticity equations.

Pour répondre au besoin de miniaturisation des spectromètres de terrain travaillant dans l’infrarouge, l’ONERA a développé un nouveau concept baptisé MICROSPOC. Ce dispositif est un détecteur infrarouge auquel a été intégré un interféromètre à deux ondes, constituant un spectromètre statique par transformée de Fourier. Ce plan focal infrarouge modifié, qui fusionne la fonction interférométrique et la fonction de détection, associé à une optique de tête simplifiée, permet d’envisager la réalisation d’instruments très compacts. L’objectif de cette thèse est de concevoir un spectromètre infrarouge miniature basé sur le dispositif MICROSPOC. Dans un premier temps, un travail théorique a été mené, dans l’objectif de dimensionner un système optique très compact. Notre choix s’étant orienté vers un système optique de collection (le détecteur voit une source étendue à distance finie), l’étude de l’acceptance angulaire de MICROSPOC dans ces conditions d’éclairement est indispensable afin de prévoir le contraste et la forme des franges d’interférence. Les résultats montreront la grande acceptance angulaire de MICROSPOC.Dans un second temps, un démonstrateur basé sur un composant MICROSPOC et sur le système optique simplifié a été réalisé. Ce démonstrateur a été caractérisé en laboratoire puis utilisé sur le terrain lors d’une campagne de mesures. Ces différentes exploitations ont montré la robustesse de l’instrument malgré l’obtention d’interférogrammes présentant divers défauts.Dans un troisième temps, une chaîne de traitement a été développée afin d’estimer un spectre à partir d’un interférogramme obtenu à l’aide du démonstrateur. Du fait des caractéristiques intrinsèques de MICROSPOC, la transformée de Fourier n’est pas la meilleure solution pour estimer un spectre. Nous l’avons montré en nous intéressant aux effets des disparités de longueur d’onde de coupure du détecteur sur l’estimation d’un spectre. Nous nous sommes alors tournés vers une approche consistant à utiliser la caractérisation spectrale de l’instrument pour inverser la mesure. Cette approche donne des résultats satisfaisants.Enfin, le but principal de cette thèse a été élargi par la conception et la réalisation de différents démonstrateurs combinant une fonction d’imagerie à une fonction de spectrométrie. Les premières pistes pour la conception d’un spectromètre qui tient dans la main ont été données. / In order to satisfy the need for handheld infrared spectrometers, the ONERA developed a new concept called MICROSPOC. This device is an infrared focal plane array with a built-in two-wave wedge-like interferometer and forms a static Fourier-transform spectrometer. This modified focal plane array, which merges the detection function and the interferometric function, in association with a simplified optical system, allows to consider the realisation of a much compact instrument. The goal of this thesis is to design and to realize a miniaturized infrared spectrometer based on the MICROSPOC concept.Firstly, a theoritical work has been led in order to design a compact optical system. Since we have chosen a collection optical system (the focal plane array sees an extended source placed at a finite distance), the study of MICROSPOC angular acceptance in these lightening conditions is needed in order to predict the contrast and the shape of interference fringes. The huge angular acceptance of MICROSPOC will be established with the results of this study.Secondly, a demonstrator based on MICROSPOC device and on the simplified optical system has been realized. This demonstrator has been caracterized in the laboratory and used in real conditions of a measurement campaign. These different exploitations have shown the robustness of the instrument despite some defaults on acquired interferograms.Then, a processing chain has been developed in order to estimate a spectrum from an interferogram acquired with our demonstrator. Considering the MICROSPOC’s own characteristics, the Fourier-transform is not the best way to estimate a spectrum. We have come to this conclusion by studying the effects of cut-off wavelenghts disparities of the detector on the spectrum estimation. At this point we have considered an approach that consists of using the spectral characterization of the instrument in order to inverse the measure. This approach gives satisfying results.Finaly, the main goal has been widened with the design and the realisation of other instruments that combine a spectrometric function and a imaging function. The first elements for the design of a handheld spectrometer have been given.

No description available.

Dans le contexte général de la résolution de problèmes inverses en acoustique et traitement du signal audio les défis sont nombreux. Pour la résolution de ces problèmes, leur caractère souvent mal posé nécessite de considérer des modèles de signaux appropriés. Les travaux de cette thèse montrent sur la base d'un cadre algorithmique générique polyvalent comment les différentes formes de parcimonie (à l'analyse ou à la synthèse, simple, structurée ou sociale) sont particulièrement adaptées à la reconstruction de signaux sonores dans un cadre mono ou multicanal. Le cœur des travaux de thèse permet de mettre en évidence les limites des conditions d'évaluation de l'état de l'art pour le problème de désaturation et de mettre en place un protocole rigoureux d'évaluation à grande échelle pour identifier les méthodes les plus appropriées en fonction du contexte (musique ou parole, signaux fortement ou faiblement dégradés). On démontre des améliorations de qualité substantielles par rapport à l'état de l'art dans certains régimes avec des configurations qui n'avaient pas été précédemment considérées, nous obtenons également des accélérations conséquentes. Enfin, un volet des travaux aborde la localisation de sources sonores sous l'angle de l'apprentissage statistique « virtuellement supervisé ». On montre avec cette méthode des résultats encourageants sur l'estimation de directions d'arrivée et de distance. / Today's challenges in the context of audio and acoustic signal processing inverse problems are multiform. Addressing these problems often requires additional appropriate signal models due to their inherent ill-posedness. This work focuses on designing and evaluating audio reconstruction algorithms. Thus, it shows how various sparse models (analysis, synthesis, plain, structured or “social”) are particularly suited for single or multichannel audio signal reconstruction. The core of this work notably identifies the limits of state-of-the-art methods evaluation for audio declipping and proposes a rigourous large-scale evaluation protocol to determine the more appropriate methods depending on the context (music or speech, moderately or highly degraded signals). Experimental results demonstrate substantial quality improvements for some newly considered testing configurations. We also show computational efficiency of the different methods and considerable speed improvements. Additionally, a part of this work is dedicated to the sound source localization problem. We address it with a “virtually supervised” machine learning technique. Experiments show with this method promising results on distance and direction of arrival estimation.

As equações diferenciais impulsivas (EDIs) modelam fenômenos que são contínuos por partes, isto é, que evoluem continuamente mas em certos momentos sofrem mudanças abruptas (impulsos) consideradas instantâneas quando comparadas à duração total do processo. Surgem nas mais diversas áreas das ciências, como na modelagem de concentração de medicamentos no corpo humano e no impacto de propaganda nas vendas de uma empresa. O problema direto associado a uma EDI com instantes de impulsos pré-fixados consiste em, fornecidos a equação diferencial, a condição inicial, os momentos de impulso e os saltos, determinar a solução do problema. Por outro lado, as incógnitas dos problemas inversos associados são os saltos e/ou os momentos de impulso. Em geral, os problemas inversos não podem ser resolvidos diretamente por meio de técnicas convencionais. A abordagem funcional é uma alternativa baseada na minimização de um funcional de erro que confronta dados do fenômeno real e do modelo matemático. O mínimo global deste funcional corresponde à solução do problema inverso. O objetivo principal desta dissertação é investigar os problemas inversos de identificação dos parâmetros saltos e momentos de impulso. Buscamos descrever uma técnica que permita tratar de problemas inversos associados às EDIs de forma bem geral, que não utilize informações específicas da aplicação além das medidas no tempo inicial e final do processo. Para isso, desenvolvemos um programa computacional composto por uma função para solução numérica do problema direto usando o método de Runge-Kutta de quarta ordem, função esta que é chamada diversas vezes para cada resolução do problema direto com diferentes valores para as incógnitas; e pelo método de otimização Simulated Annealing que altera sistematicamente os valores das incógnitas. Os resultados mostram que resolver os problemas inversos que surgem das EDIs não é uma tarefa simples, que a técnica estudada é promissora e que pode ser aperfeiçoada / Impulsive differential equations (IDEs) model piecewise continuous phenomena, that is, that evolve continuously but at certain moments suffer abrupt changes (impulses) considered instantaneous when compared to the total duration of the process. They arise in several areas of science, such as the modeling of drug concentration in the human body and the impact of advertising on a companys sales. The direct problem associated with an IDE with impulses at fixed times consists of determining the solution to the problem, provided the differential equation, the initial condition, the moments of impulse and the jumps. On the other hand, the unknowns of the associated inverse problems are the jumps and/or the moments of impulse. In general, the inverse problems cannot be solved directly by conventional techniques. The functional approach is an alternative based on the minimization of an error functional that confronts data of the real phenomenon and the mathematical model. The global minimum of this functional corresponds to the solution of the inverse problem. The main objective of this dissertation is to investigate the inverse problems of jumps and moments of impulse parameters identification. We have attempted to describe a technique that allows treating of the inverse problems associated with IDEs in a general way, which does not use particular information of the application besides the measurements in the initial and final time of the process. For this, we developed a computer program composed by a function to solve the direct problem numerically using the fourth-order Runge-Kutta method, which is called several times for each resolution of the direct problem with different values for the unknowns; and by the Simulated Annealing optimization method which changes the values of the unknowns systematically. The results show that solving the inverse problems that arise from IDEs is not a simple task and that the technique studied is promising and can be improved

Técnicas de reconstrução térmica inversa são muito usadas em diferentes aplicações tais como a determinação de propriedades térmicas de novos materiais, controle da produção de calor, temperatura em processos de manufatura, etc. Apesar da ampla aplicabilidade, o problema inverso é intrinsecamente mal condicionado e tem sido tema de trabalhos de vários pesquisadores. A solução de um problema térmico inverso tridimensional é significantemente complexa, e, assim requer uma formulação que não contenha condições experimentais não realistas tais como confinamento bidimensional e estabilidade do campo térmico com relação a mudanças em parâmetros internos. Uma das abordagens adotada é baseada na formulação variacional sobre a forma do erro quadrático para reconstrução da distribuição de condução de calor interna e coeficiente de condução de calor parietal para um problema tridimensional. Dentro desta estrutura, a natureza mal condicionada do problema se manifesta na superfície de otimização por produzir topologias problemáticas tais como, vários mínimos locais, pontos de sela, vales e platôs ao redor da solução etc. Para viabilizar a abordagem escolhida, um modelo numérico foi escrito baseado na discretização por diferenças finitas da equação diferencial governante e condições de contorno. O erro funcional foi definido pela comparação entre medidas experimentais e numéricas de temperatura. O objetivo foi realizar simulações numéricas a fim de mapear a superfície de otimização correspondente e identificar a estrutura problemática associada ou patologia, chegando assim à reconstrução do coeficiente de convecção h. / Inverse thermal reconstruction techniques are widely used in different applications such as the determination of thermal properties of new materials, control of heat generation, temperature in manufacturing processes, etc. Despite the broad range of applicability, an inverse problem is intrinsically ill conditioned and has been the subject of the work of several researchers. The solution of an inverse 3-dimesional thermal problem is significantly complex, and, thus, requires a formulation that do not contain unrealistic experimental conditions such as 2-dimensional confinement and steadiness of the thermal field with respect to changes in internal parameters. One of the most adopted is the variational formulation based on quadratic error forms for the reconstruction of the internal heat conduction distribution and convection coefficient for a 3-dimensional problem. Within this framework, the ill conditioned nature of the problem manifests itself on the optimization surface by producing problematic topologies such as contour and multiple local minima, saddle points, plateaux around the solution pit and so on. To be able to apply th method a numerical model was written based on a finite difference discretization of the governing differential equation and boundary conditions. An error functional was defined by comparing experimental and numerical measurement temperatures. Numerical simulations aiming at mapping the corresponding optimization surfaces andatidentifing the associated problematic structures or pathologies, resulting in the reconstruction of convection coefficient.

A identificação de esforços dinâmicos vem despertando um contínuo interesse da comunidade modal e de teste de vibrações principalmente nas duas últimas décadas. De forma semelhante como na maioria das técnicas de identificação modal, a identificação de forças, representa um problema inverso. Problemas inversos são conhecidos pelo fato de serem naturalmente mau condicionados numericamente, uma vez que os mesmos representam uma operação de deconvolução no tempo. Dentre os métodos utilizados na identificação de forças a partir de medidas de movimento, pode-se destacar a técnica da pseudo inversa no domínio da frequência. A técnica da pseudo inversa requer o conhecimento das medidas das respostas da estrutura usualmente acelerações assim como das funções repostas em frequência da estrutura (FRF). Este método requer a inversão das matrizes de FRF dos sistemas para todas as linhas espectrais na faixa de frequência de interesse. Este procedimento de inversão usualmente apresenta dificuldades numéricas em algumas frequências. Será levantado suas principais vantagens e desvantagens quando comparado com um método no domínio do tempo no caso da técnica da SWAT, para alguns tipos de formas de excitação comumente empregada sem testes modais de vibração. São apresentados resultados de simulações numéricas e resultados experimentais para uma estrutura simples. / The main goal of this dissertation is to takle a difficult problem in experimental modal analysis that is the indirect identification of input forces based on the knowledge of the structure\'s output response and dynamic characteristics, such as impulse responses and/or FRF. This represents an inverse problem in mechanics and usually offers great numerical difficulties in the process of identification of the input forces. There are methods for input force identification in the lime and in the frequency domains. In lhe time domain the SWAT method (Sum of lhe Weighet Accelerations Technique) is a method that is based on the principle of motion of the mass center. This technique alows the estimation of the resulting force acting on the structure and is primarily used with impact force signals. In the frequency domain the pseudo-inverse technique is known to give estimates of the resulting forces acting on the structure under investigation. In this work numerically simulated as well as experimental results are presented for both methods and their major advantages and disadvantages are discussed. The pseudo-inverse method is further employed in the identification of experimental transient and random multiple inputs.