Search

Artículo escrito por un elevado número de autores, solo se referencian el que aparece en primer lugar, los autores pertenecientes a la UAM y el nombre del grupo de colaboración, si lo hubiere, We present the design, fabrication, and characterization of an implantable neural interface based on anisotropic magnetoresistive (AMR) magnetic-field sensors that combine reduced size and high performance at body temperature. The sensors are based on La0.67Sr0.33MnO3 (LSMO) as a ferromagnetic material, whose epitaxial growth has been suitably engineered to get uniaxial anisotropy and large AMR output together with low noise even at low frequencies. The performance of LSMO sensors of different film thickness and at different temperatures close to 37 °C has to be explored to find an optimum sensitivity of ∼400%/T (with typical detectivity values of 2 nT·Hz-1/2 at a frequency of 1 Hz and 0.3 nT·Hz-1/2 at 1 kHz), fitted for the detection of low magnetic signals coming from neural activity. Biocompatibility tests of devices consisting of submillimeter-size LSMO sensors coated by a thin poly(dimethyl siloxane) polymeric layer, both in vitro and in vivo, support their high suitability as implantable detectors of low-frequency biological magnetic signals emerging from heterogeneous electrically active tissues

Thanks to its ability to provide dense strain rate measurements along Optical Fiber (OF) cable, the Distributed Acoustic Sensing (DAS) technique spreads over different seismic and geophysical domains. They range from exploration geophysics (Mestayer et al. 2011, Daley et al. 2013), to underground structure imaging (e.g. Ajo-Franklin et al. 2019, Cheng et al. 2021) or seismic activity and background noise monitoring (Jousset et al 2018, Nayak et al. 2021). Beyond the advantage of its dense spatial sampling and given a better understanding of its instrument response (e.g. Lindsey et al. 2020), the detection performance of these new DAS measurements also depends on its ability to precisely characterize the amplitude and phase of the seismic background noise in different environments. According to recent offshore seismic noise studies (Ugalde et al. 2021, Lior et al. 2021, Guerin et al 2022), we propose a study based on DAS recordings of the seismic background noise in an on-land quiet environment.In 2020, a temporary seismic experiment PREMISE (PREliminary MIga Seismic Experiment) was carried out on the site of the underground low noise Laboratory (LSBB, Laboratoire Souterrain Bas Bruit) at Rustrel, France, in order to study the 3D seismic wave field properties in a pretty well-known underground geological structure. During this experiment, we deployed several kilometers of different OF in the LSBB galleries in order to create a multidirectional DAS array with a total fiber length of 10.5km and several ground-coupling conditions. We reprocessed two hours of “raw” DAS data, recorded with a FEBUS A1-R instrument, with different acquisition parameters to find the best configuration for enhancing the DAS measurement Signal to Noise Ratio. The power spectral density (PSD) of these reprocessed strain time-series reveals a peak in the background noise frequency range [0.08-0.25Hz] for gauge lengths of 90m and 150m. Independently, an estimation of the local strain field has been derived by a geodetic analysis (Spudich et al 1995) of the records from the LSBB broadband seismometers antenna. The comparison of the DAS and seismometers array-derived strain PSD shows a very good agreement with the secondary microseism peak in terms of frequency band, amplitude, and the wave field polarization, especially for DAS strain records processed with gauge-length of several tens of meters.

International audience; The current trend in magnetoresistive sensors development is to increase the sensitivity of single sensing elements by using multilayer structures and to design them into arrays. Such arrays are designed to compensate the excess low frequency noise of individual elements, which limits their magnetic resolution. Here, we report the modeling, design, and fabrication of single layer anisotropic magnetoresistive (AMR) sensors using low noise epitaxial La2/3Sr1/3MnO3 (LSMO) oxide thin films deposited on vicinal SrTiO3 substrates. The fabrication process is simple, and the operation of the sensor is based on a step-induced uniaxial magnetic anisotropy, described using the Stoner–Wohlfarth model. A coherent magnetization reversal process is observed by magneto-optical Kerr effect imaging. A good agreement between experimental data and the expected sensor response confirms the correct operation of the device. Three main fabrication parameters, namely the vicinal angle of the substrate, the deposition temperature, the thin film thickness, and their effects on film anisotropy field and device detectivity have been studied. Detectivity levels as low as 1.4 nT Hz–1/2 at 1 Hz and 240 pT Hz–1/2 in the white noise region are achieved with a single Wheatstone bridge element operating at 310 K. Compared to GMR and AMR sensors, these results are promising for further development and for their use as single layer LSMO low field AMR sensors, including applications as implantable biomedical devices.

The yeast Saccharomyces cerevisiae is powerful for studying human G protein-coupled receptors as they can be coupled to its mating pathway. However, some receptors, including the mu opioid receptor, are non-functional, which may be due to the presence of the fungal sterol ergosterol instead of cholesterol. Here we engineer yeast to produce cholesterol and introduce diverse mu, delta, and kappa opioid receptors to create sensitive opioid biosensors that recapitulate agonist binding profiles and antagonist inhibition. Additionally, human mu opioid receptor variants, including those with clinical relevance, largely display expected phenotypes. By testing mu opioid receptor-based biosensors with systematically adjusted cholesterol biosynthetic intermediates, we relate sterol profiles to biosensor sensitivity. Finally, we apply sterol-modified backgrounds to other human receptors revealing sterol influence in SSTR5, 5-HTR4, FPR1, and NPY1R signaling. This work provides a platform for generating human G protein-coupled receptor-based biosensors, facilitating receptor deorphanization and high-throughput screening of receptors and effectors.

Dans un secteur rural au nord-est de la France, une collectivité locale souhaite relancer son économie en développant un projet de tourisme proposant diverses approches d’éducation relative à l’environnement, centrées sur l’environnement-nature. Pour la conception de ce « Pôle Nature », des personnes aux intérêts parfois très divergents (élus, chasseurs, pêcheurs, agriculteurs, naturalistes, scientifiques, éducateurs, etc.) ont été sollicitées. Dans le cadre de cet article, nous souhaitons montrer que les principales réserves et réticences exprimées sur ce projet peuvent s’interpréter par la présence d’un conflit de représentations relatives aux pratiques touristiques entre des acteurs ayant des rapports à la nature différents. In a rural sector in the North-East of France, a local community wishes to promote its economy by developing a tourism project that proposes different environmental education approaches based on the environment as nature. For the design of this « Pôle Nature », people with sometimes very divergent interests (elected officials, hunters, fishermen, farmers, naturalists, scientists, teachers, etc.) were solicited. Within this paper, we wish to show that the principal reserves expressed on this project can be explained by the presence of a conflict on the representations of tourism practices between the actors that have different relationships with nature.

International audience; We report the integration of high-quality epitaxial La 2/3 Sr 1/3 MnO 3 (LSMO) thin films onto SrTiO 3 buffered Silicon-on-Sapphire (SOS) substrates by combining state-of-the-art thin film growth techniques such as molecular beam epitaxy and pulsed laser deposition. Detailed structural, magnetic and electrical characterizations of the LSMO/STO/SOS heterostructures show that the LSMO film properties are competitive with those directly grown on oxide substrates. X-ray magnetic circular dichroism measurements on Mn L 2,3 edges show strong dichroic signal at room temperature, and angular-dependent in-plane magnetic properties by magneto-optical Kerr magnetometry reveal isotropic magnetic anisotropy. Suspended micro-bridges were thus finally fabricated by silicon micromachining, thus demonstrating the potential use of integrating LSMO magnetic layer on industrially compatible SOS substrates for the development of applicative MEMS devices.

A formal [4+1]-annulation strategy between sulfur ylides and 1,3-dienes was developed to afford functionalized cyclopentanoids. The process consists of a stereoselective cyclopropanation reaction followed, in situ, by a stereospecific MgI2 -catalyzed vinylcyclopropane-cyclopentene rearrangement. The use of chiral sulfur ylides provided cyclopentanoids with excellent enantiocontrol. A combined experimental and computational mechanistic study showed that the stereospecificity of the rearrangement could be accounted for by a double SN 2 reaction mechanism involving iodide.

Ce texte montre comment une institution patrimoniale peut participer à l’interprétation du territoire et à l’éducation au développement durable. Il interroge, notamment, la notion d’appropriation territoriale dans le champ de l’éducation aux territoires en contexte muséal, mettant ainsi en lumière la complexité des dynamiques de construction identitaire et de développement local., This text shows how a heritage institution can participate in the interpretation of the territory and in education for sustainable development. In particular, it questions the notion of territorial appropriation in the field of education to territories in a museum context, thus highlighting the complexity of the dynamics of identity construction and local development.

International audience; Image compression at very low bitrate has known a new dawn since the democratization of convolutional neural networks (CNNs). A promising approach consists in using an « end-to-end » neural network-based compression scheme, performing both coding and decoding phases. Such techniques lead to perceptually convincing results, notably with GAN-based architectures. However, due to the nature of GAN, alongside being computationally expensive on the encoder side, such schemes cannot ensure the preservation of fine details, such as small digits or letters on an ID card. To overcome this limitation, we propose a new model, specifically trained to reduce artifacts resulting from strong lossy compressions. An advantage of the proposed approach comes from the fact that it can be interpreted as a post-processing step that can easily be added to any compression scheme without modifying the codec. Experimental results show that our model perceptually outperforms the state-of-the-art compression standards for very low bitrates.

International audience; The problem of data storage and transmission on mobile devices is constantly growing up. Smartphones are nearly by default equipped with HD cameras that are taking high quality pictures, which can be instantly stored and easily uploaded on cloud platforms. Such a behavior favors the creation of a massive amount of data. In order to reduce the size of such data, it is mandatory to dispose of efficient compression techniques that can take into account the actual usage of such image data. For example, most of the pictures acquired by phone cameras are often displayed on a small screen and this, for a little amount of time. A solution to manage this kind of oversized data would be to store them in a lower resolution, additionally to a standard image compression. The downside of such an approach is that restoring an image to its original resolution is a challenging task, notably in the presence of complex compression artifacts, such as those introduced by sophisticated compression methods. In order to deal with such an issue, in this paper we propose a new model, specifically trained to perform super-resolution after compression with the BPG state-of-the-art codec. An advantage of the proposed approach comes from the fact that the underlying process can be interpreted as a postprocessing step, which can be easily added to any compression scheme without modifying the codec. Experimental results show that our model perceptually outperforms the state-of-the-art compression standards even for very low bitrates.

Through the application of a region-focused saturation mutagenesis and randomization approach, protein engineering of the Cal-A enzyme was undertaken with the goal of conferring new triglyceride selectivity. Little is known about the mode of triglyceride binding to Cal-A. Engineering Cal-A thus requires a systemic approach. Targeted and randomized Cal-A libraries were created, recombined using the Golden Gate approach and screened to detect variants able to discriminate between long-chain (olive oil) and short-chain (tributyrin) triglyceride substrates using a high-throughput in vivo method to visualize hydrolytic activity. Discriminative variants were analyzed using an in-house script to identify predominant substitutions. This approach allowed identification of variants that exhibit strong discrimination for the hydrolysis of short-chain triglycerides and others that discriminate towards hydrolysis of long-chain triglycerides. A clear pattern emerged from the discriminative variants, identifying the 217-245 helix-loop-helix motif as being a hot-spot for triglyceride recognition. This was the consequence of introducing the entire mutational load in selected regions, without putting a strain on distal parts of the protein. Our results improve our understanding of the Cal-A lipase mode of action and selectivity. This holistic perspective to protein engineering, where parts of the gene are individually mutated and the impact evaluated in the context of the whole protein, can be applied to any protein scaffold.

The synthesis of various electron-deficient 1,1-dicarbon- yl ester 1,3-dienes substituted in position 2 or 3 of the diene moiety has been developed.

Functionalized cyclopentenes are conveniently obtained via stereoselective cyclopropanation reaction of sulfur ylides with 1,2-dienes followed by in situ stereospecific MgI2-promoted rearrangement of vinylcyclopropanes.

SummaryVarious antiphospholipid and/or antiprotein antibodies have been suspected to be associated with recurrent early foetal loss in absence of any habitual aetiology. We conducted a hospital-based case control study on women with no antecedent of thromboembolic or autoimmune disease. We studied 3 groups of 518 women: patients with unexplained primary recurrent early foetal loss, patients with explained episodes and mothers with no previous obstetrical accident. Matching the 3 groups was carried out on the basis of age, number or pregnancies and time elapsed since the end of the last pregnancy. Significant biological markers were then prospectively tested.The various antibodies were shown to be dependent on parity and on the presence of previous foetal loss: cut-off values were thus calculated using data obtained from the group of explained accidents, and adjusted for parity. Only anti-phosphatidylethanolamine IgM [odds ratio: 6.0, 95% confidence interval (2.3–15.7), p = 0.0003], anti-β2-glycoprotein I IgG [4.4, (1.6–11.7), p = 0.0035] anti-annexin V IgG antibodies [3.2 (1.2–8.1), p = 0.015] and lupus anticoagulant [3.0, (1.3–6.8), p = 0.009], were found to be independent retrospective risk factors for unexplained early foetal loss. These four markers were subsequently found to be, during the following pregnancy, associated with a significant risk of foetal loss despite a low-dose aspirin treatment.In non-thrombotic, non-auto-immune women with unexplained primary reccurent early foetal loss, subgroups of patients with positive anti-phosphatidylethanolamine IgM antibodies, or positive anti- β2-glycoprotein-I IgG antibodies, or positive anti-annexin V IgG anti-bodies or lupus anticoagulant must be particularised. This should allow therapeutic trials to be carried in well-defined patients.

We report an innovative label-free biosensor based on statistical analysis of several whispering gallery modes spectral shifts in polystyrene fluorescent microspheres using a custom microflow cytometer. Whispering gallery modes analysis enables detection of nanometer-sized analytes showing promising possibilities for virus, bacteria and molecular detection. To demonstrate this, fluorophore-doped microspheres of the appropriate size parameter are mixed in an aqueous solution. Then, a syringe pump pushes the solution through a fiber optic flow cell where a laser beam illuminates the analysis area to excite the microspheres and their fluorescence is collected. This device provides a low-cost and user friendly solution that could enhance spectrum acquisition rates up to 5 spectra per second thanks to the considerable amount of microspheres flowing through the excitation area per unit time. Finally, the fluorescence spectra are statistically investigated using an instantaneous measurement of apparent refractive index algorithm to determine a reliable value for the refractive index of the environment since the exact radius of the microsphere scanned is unknown. This refractive index becomes an effective value for the local perturbation caused by inhomogeneities on the microsphere surface and hence, determines whether or not inhomogeneities, such as bacteria, are adsorbed by comparing to a control sample. Combining a flow cell with our detection algorithm, we reduce the period of a 50 microspheres experiment from 161 minutes to 14 minutes when the flow rate is 2000 µl/h and the microsphere concentration is 5 µsphere/µl.

La révolution des communications est si profonde qu'elle transforme notre rapport au réel, notre mode d'acquisition de l'information, notre relation à la culture. Nous ne sommes plus les héritiers d'une civilisation bien balisée, nous sommes des pionniers qui entrons dans une ère nouvelle. Dans le texte ci-dessous, Philippe-Olivier Rousseau, membre du Conseil supérieur de l'audiovisuel, précise ce que sont les nouvelles technologies de l'information, leurs prouesses et leurs pièges, mais il élève en même temps le débat pour souligner les défis qu'elles lancent à l'homme., Rousseau Philippe-Olivier. Pour un nouvel encyclopédisme : quelques paradoxes à propos des technologies de l'information. In: Communication et langages, n°100-101, 2ème-3ème trimestre 1994. 25e anniversaire. pp. 142-152.

This article proposes two avenues to help improve the realism of numerical computations for cardiac electrophysiology while maintaining manageable computational resources. We first propose an asymptotic analysis to adjust the parameters and use a simple two-variable ionic model to reproduce the main characteristics of the cardiac action potential (AP) in various myocardial regions. This ionic model is embedded in the bidomain model that is used to propagate the AP in the heart. Our second contribution is a finite element method that couples the heart with the torso in a single variational formulation and allows non body fitted meshes at the interface between the myocardium and the torso/ventricle cavities. This interface is described through a level-set function obtained from the segmentation of patient medical images. Using a 2D test case, we compare the use of body-fitted and non body-fitted meshes and analyze the impact of both approaches on the accuracy of the solutions, including an anisotropic mesh adaptation strategy.

International audience; This paper presents a 2D/3D discrete duality finite volume method for solving heterogeneous and anisotropic elliptic equations on very general unstructured meshes. The scheme is based on the definition of discrete divergence and gradient operators that fulfill a duality property mimicking the Green formula. As a consequence, the discrete problem is proved to be well-posed, symmetric and positive-definite. Standard numerical tests are performed in 2D and 3D and the results are discussed and compared with P1 finite elements ones. At last, the method is used for the resolution of a problem arising in biomathematics: the electrocardiogram simulation on a 2D mesh obtained from segmented medical images.

It has been suggested the risk of hydrocarbon-induced chronic nephropathy is negligible at low exposure levels. The first purpose of the study was to test this hypothesis by selecting a population slightly exposed to hydrocarbons. Moreover, as hypertension might be associated with an increased excretion of nephrotoxic mercapturates, the association between blood pressure and urinary concentration of S-phenylmercapturic acid (S-PMA) was also examined. Lifetime exposure assessment, main tests of subclinical kidney damage, and statistical approach were taken from a previous study that had included primarily moderately or heavily exposed workers and had found hydrocarbon-induced nephrotoxic effects. No nephrotoxic effect of exposure could be ascertained in the present study. S-PMA concentration was not increased in hypertensive workers. Thus, the risk of hydrocarbon-induced chronic nephropathy might be extremely low in workers slightly exposed to hydrocarbons. The negative results of some studies might be due to the low lifetime hydrocarbon exposures of the study populations.

The absolute configuration of the four stereoisomeric α-substituted derivatives of dota has been defined; in their Eu complexes, 2D-NMR methods have revealed that in solution ring inversion is independent of pendant arm rotation and the (RRRR)-[EuL1a] complex which in the solid state crystallises as a square antiprismatic structure.

International audience

This engaging book addresses the question of how diverse communities, whether in a nation, city or organization, can live together and prosper whilst retaining and enjoying their cultural differences. This is a particularly pertinent issue in the context of the modern world where mass migration and immigration are pervasive global phenomena.

The field of visual content compression (image, video, 2D/3D graphics elements) has known spectacular achievements for more than twenty years, with the emergence numerous international standards such as JPEG, JPEG2000 for still image compression, or MPEG-1/2/4 for video and 3D graphics content coding.The apparition of smartphones and of their related applications have also benefited from these advances, the image being today ubiquitous in a context of mobility. Nevertheless, image transmission requires reliable and available networks, since such visual data that are inherently bandwidth-intensive. While developed countries benefit today from high-performance mobile networks (3G, 4G...), this is not the case in a certain number of regions of the world, particularly in emerging countries, where communications still rely on 2G SMS networks. Transmitting visual content in such a context becomes a highly ambitious challenge, requiring the elaboration of new, for very low bitrate compression algorithm. The challenge is to ensure images transmission over a narrow bandwidth corresponding to a relatively small set (10 to 20) of SMS (140 bytes per SMS).To meet such constraints, multiple axes of development have been considered. After a state-of-the-art of traditional image compression techniques, we have oriented our research towards deep learning methods, aiming achieve post-treatments over strongly compressed data in order to improve the quality of the decoded content.Our contributions are structures around the creation of a new compression scheme, including existing codecs and a panel of post-processing bricks aiming at enhancing highly compressed content. Such bricks represent dedicated deep neural networks, which perform super-resolution and/or compression artifact reduction operations, specifically trained to meet the targeted objectives. These operations are carried out on the decoder side and can be interpreted as image reconstruction algorithms from heavily compressed versions. This approach offers the advantage of being able to rely on existing codecs, which are particularly light and resource-efficient. In our work, we have retained the BPG format, which represents the state of art in the field, but other compression schemes can also be considered.Regarding the type of neural networks, we have adopted Generative Adversarials Nets-GAN, which are particularly well-suited for objectives of reconstruction from incomplete data. Specifically, the two architectures retained and adapted to our objectives are the SRGAN and ESRGAN networks. The impact of the various elements and parameters involved, such as the super-resolution factors and the loss functions, are analyzed in detail.A final contribution concerns experimental evaluation performed. After showing the limitations of objective metrics, which fail to take into account the visual quality of the image, we have put in place a subjective evaluation protocol. The results obtained in terms of MOS (Mean Opinion Score) fully demonstrate the relevance of the proposed reconstruction approaches.Finally, we open our work to different use cases, of a more general nature. This is particularly the case for high-resolution image processing and for video compression.; Le domaine de la compression de contenus visuels (image, vidéo, éléments graphiques 2D/3D) a connu, depuis maintenant plus de vingt ans, un essor considérable avec l’émergence notamment au fil des années de nombreuses normes internationales comme JPEG, JPEG2000 pour les images fixes ou les différentes versions de standards MPEG-1/2/4 pour les données vidéo et graphiques.L’apparition des smartphones et l’explosion des applications qui leur sont dédiées a également bénéficié de ces avancées, l’image étant aujourd’hui omniprésente dans un contexte de mobilité/itinérance. Néanmoins, cela nécessite toujours des réseaux fiables et disponibles, offrant un débit suffisant pour la transmission effective de ces données visuelles qui sont intrinsèquement gourmandes en bande passante. Si aujourd’hui les pays développés bénéficient de réseaux mobiles (3G, 4G…) hautement performantes, cela n’est pas le cas d’un certain nombre de régions du monde, en particulier dans les pays émergents, où les communications s’appuient encore sur des réseaux 2G SMS. Transmettre de contenus visuels dans un tel contexte devient un défi ambitieux, qui nécessite la mise en œuvre de nouveaux algorithmes de compression. Le défi à relever consiste à assurer une transmission des images sur une bande passante correspondant à un ensemble relativement réduit (10 à 20) de SMS (140 octets par SMS).Pour répondre à ces contraintes, de multiples pistes de développement ont été envisagées. Après un état de l’art des techniques de compression traditionnelles et de leurs améliorations futures, nous avons finalement orienté nos travaux vers des méthodes de deep learning, visant à réaliser des post-traitements pour améliorer la qualité des contenus compressés.Nos contributions s’articulent autour de la création d’un nouveau schéma de compression, incluant les codecs existants ainsi qu’un panel de briques de post-traitement permettant une meilleure exploitation des contenus fortement compressés. Ces briques sont des réseaux de neurones profonds dédiés, qui réalisent des opérations de super-résolution et/ou de réduction d’artéfacts de compression, spécifiquement entraînés pour répondre aux objectifs ciblés. Ces opérations interviennent du côté du décodeur et peuvent être interprétées comme des algorithmes de reconstruction d’images à partir de versions fortement compressées. Cette approche présente l’avantage de pouvoir s’appuyer des codecs existants, particulièrement légers et peu coûteux en ressources. Dans nos travaux, nous avons retenu le format BPG, qui fait état de l’art dans le domaine, mais d’autre schémas de compression peuvent être également considérés.Concernant le type de réseaux de neurones, nos recherches nous ont conduits vers les réseaux antagonistes génératifs (Generative Adversarials Nets–GAN), qui s‘avèrent particulièrement adaptés pour des objectifs de reconstruction à partir de données incomplètes. Plus précisément, les deux architectures retenues et adaptées à nos objectifs sont les réseaux SRGAN et ESRGAN. L’impact des différents éléments et paramètres impliqués, comme notamment les facteurs de super-résolution utilisés et les fonctions de pertes, sont analysés en détails.Enfin, une dernière contribution concerne l’évaluation expérimentale. Après avoir montré les limitations des métriques objectives, qui peinent à prendre en compte la qualité visuelle de l’image, nous avons mis en place un protocole d’évaluation subjective. Les résultats obtenus en termes de scores MOS (Mean Opinion Score) démontrent pleinement la pertinence des approches de reconstruction proposées.Enfin, nous analysons une ouverture de nos travaux à des cas d’utilisation différents, d’une nature plus grand public. C’est notamment le cas pour le traitement de contenus de grande résolution plus ou moins compressés et même pour l’amélioration de la qualité de vidéos.

The field of visual content compression (image, video, 2D/3D graphics elements) has known spectacular achievements for more than twenty years, with the emergence numerous international standards such as JPEG, JPEG2000 for still image compression, or MPEG-1/2/4 for video and 3D graphics content coding.The apparition of smartphones and of their related applications have also benefited from these advances, the image being today ubiquitous in a context of mobility. Nevertheless, image transmission requires reliable and available networks, since such visual data that are inherently bandwidth-intensive. While developed countries benefit today from high-performance mobile networks (3G, 4G...), this is not the case in a certain number of regions of the world, particularly in emerging countries, where communications still rely on 2G SMS networks. Transmitting visual content in such a context becomes a highly ambitious challenge, requiring the elaboration of new, for very low bitrate compression algorithm. The challenge is to ensure images transmission over a narrow bandwidth corresponding to a relatively small set (10 to 20) of SMS (140 bytes per SMS).To meet such constraints, multiple axes of development have been considered. After a state-of-the-art of traditional image compression techniques, we have oriented our research towards deep learning methods, aiming achieve post-treatments over strongly compressed data in order to improve the quality of the decoded content.Our contributions are structures around the creation of a new compression scheme, including existing codecs and a panel of post-processing bricks aiming at enhancing highly compressed content. Such bricks represent dedicated deep neural networks, which perform super-resolution and/or compression artifact reduction operations, specifically trained to meet the targeted objectives. These operations are carried out on the decoder side and can be interpreted as image reconstruction algorithms from heavily compressed versions. This approach offers the advantage of being able to rely on existing codecs, which are particularly light and resource-efficient. In our work, we have retained the BPG format, which represents the state of art in the field, but other compression schemes can also be considered.Regarding the type of neural networks, we have adopted Generative Adversarials Nets-GAN, which are particularly well-suited for objectives of reconstruction from incomplete data. Specifically, the two architectures retained and adapted to our objectives are the SRGAN and ESRGAN networks. The impact of the various elements and parameters involved, such as the super-resolution factors and the loss functions, are analyzed in detail.A final contribution concerns experimental evaluation performed. After showing the limitations of objective metrics, which fail to take into account the visual quality of the image, we have put in place a subjective evaluation protocol. The results obtained in terms of MOS (Mean Opinion Score) fully demonstrate the relevance of the proposed reconstruction approaches.Finally, we open our work to different use cases, of a more general nature. This is particularly the case for high-resolution image processing and for video compression., Le domaine de la compression de contenus visuels (image, vidéo, éléments graphiques 2D/3D) a connu, depuis maintenant plus de vingt ans, un essor considérable avec l’émergence notamment au fil des années de nombreuses normes internationales comme JPEG, JPEG2000 pour les images fixes ou les différentes versions de standards MPEG-1/2/4 pour les données vidéo et graphiques.L’apparition des smartphones et l’explosion des applications qui leur sont dédiées a également bénéficié de ces avancées, l’image étant aujourd’hui omniprésente dans un contexte de mobilité/itinérance. Néanmoins, cela nécessite toujours des réseaux fiables et disponibles, offrant un débit suffisant pour la transmission effective de ces données visuelles qui sont intrinsèquement gourmandes en bande passante. Si aujourd’hui les pays développés bénéficient de réseaux mobiles (3G, 4G…) hautement performantes, cela n’est pas le cas d’un certain nombre de régions du monde, en particulier dans les pays émergents, où les communications s’appuient encore sur des réseaux 2G SMS. Transmettre de contenus visuels dans un tel contexte devient un défi ambitieux, qui nécessite la mise en œuvre de nouveaux algorithmes de compression. Le défi à relever consiste à assurer une transmission des images sur une bande passante correspondant à un ensemble relativement réduit (10 à 20) de SMS (140 octets par SMS).Pour répondre à ces contraintes, de multiples pistes de développement ont été envisagées. Après un état de l’art des techniques de compression traditionnelles et de leurs améliorations futures, nous avons finalement orienté nos travaux vers des méthodes de deep learning, visant à réaliser des post-traitements pour améliorer la qualité des contenus compressés.Nos contributions s’articulent autour de la création d’un nouveau schéma de compression, incluant les codecs existants ainsi qu’un panel de briques de post-traitement permettant une meilleure exploitation des contenus fortement compressés. Ces briques sont des réseaux de neurones profonds dédiés, qui réalisent des opérations de super-résolution et/ou de réduction d’artéfacts de compression, spécifiquement entraînés pour répondre aux objectifs ciblés. Ces opérations interviennent du côté du décodeur et peuvent être interprétées comme des algorithmes de reconstruction d’images à partir de versions fortement compressées. Cette approche présente l’avantage de pouvoir s’appuyer des codecs existants, particulièrement légers et peu coûteux en ressources. Dans nos travaux, nous avons retenu le format BPG, qui fait état de l’art dans le domaine, mais d’autre schémas de compression peuvent être également considérés.Concernant le type de réseaux de neurones, nos recherches nous ont conduits vers les réseaux antagonistes génératifs (Generative Adversarials Nets–GAN), qui s‘avèrent particulièrement adaptés pour des objectifs de reconstruction à partir de données incomplètes. Plus précisément, les deux architectures retenues et adaptées à nos objectifs sont les réseaux SRGAN et ESRGAN. L’impact des différents éléments et paramètres impliqués, comme notamment les facteurs de super-résolution utilisés et les fonctions de pertes, sont analysés en détails.Enfin, une dernière contribution concerne l’évaluation expérimentale. Après avoir montré les limitations des métriques objectives, qui peinent à prendre en compte la qualité visuelle de l’image, nous avons mis en place un protocole d’évaluation subjective. Les résultats obtenus en termes de scores MOS (Mean Opinion Score) démontrent pleinement la pertinence des approches de reconstruction proposées.Enfin, nous analysons une ouverture de nos travaux à des cas d’utilisation différents, d’une nature plus grand public. C’est notamment le cas pour le traitement de contenus de grande résolution plus ou moins compressés et même pour l’amélioration de la qualité de vidéos.