Portail HAL ONERA
Not a member yet
12842 research outputs found
Sort by
Multispectral Texture Synthesis using RGB Convolutional Neural Networks
No full textInternational audienceState-of-the-art RGB texture synthesis algorithms rely on style distances that are computed through statistics of deep features. These deep features are extracted by classification neural networks that have been trained on large datasets of RGB images. Extending such synthesis methods to multispectral images is not straightforward, since the pre-trained networks are designed for and have been trained on RGB images. In this work, we propose two solutions to extend these methods to multispectral imaging. Neither of them require additional training of the neural network from which the second order neural statistics are extracted. The first one consists in optimizing over batches of random triplets of spectral bands throughout training. The second one projects multispectral pixels onto a 3 dimensional space. We further explore the benefit of a color transfer operation upstream of the projection to avoid the potentially abnormal color distributions induced by the projection. Our experiments compare the performances of the various methods through different metrics. We demonstrate that they can be used to perform exemplar-based texture synthesis, achieve good visual quality and comes close to state-of-the art methods on RGB bands. Code is available at \url{https://github.com/selim2483/multispectral-textureCNN
Détection et caractérisation de la pollution par les plastiques sur les surfaces continentales
No full textInternational audienc
Comment réduire l'effet mémoire dans les caméras à microbolomètres ?
No full textInternational audienceMotion blur is frequently observed in images captured by microbolometer cameras. This is due to the memory effect of this type of sensor, which corrupts the response not only in the presence of relative motion between the camera and the target object, but also in the event of rapid variations in the temperature of a static target. From a thermal point of view, microbolometer sensors behave like first-order dynamic systems. For this reason, the memory of a past radiative event remains present in the sensor's temperature level and decays exponentially with a time scale corresponding to what is known as the sensor's thermal time constant. We propose an image processing method that eliminates, at least partially, the memory effect of microbolometer cameras. Specifically, in any image, the memory of radiative events that took place before the previous image can be completely erased. Motion blur is therefore considerably reduced. Thanks to this patented procedure, the microbolometer camera can be considered as changing from a time-constant sensor to an integration-time sensor, like cooled photon detector cameras. This specific integration time corresponds to the delay between two readout cycles. For this reason, motion blur can be further reduced by increasing the camera's readout frequency. The advantage of this method is that it does not involve any complicated deconvolution procedure and is easy to implement and real-time. Nor does it depend on the direction of the relative motion. It considers each pixel individually and removes from the current signal the influence of the incident radiation history that is older than the previous pixel reading.Le flou de bougé est fréquemment observé dans les images capturées par les caméras à microbolomètres. Cela est dû à l'effet mémoire de ce type de capteur, qui affecte la réponse non seulement en présence d'un mouvement relatif entre la caméra et l'objet cible, mais aussi en cas de variations rapides de la température d'une cible statique. D'un point de vue thermique, les capteurs microbolométriques se comportent comme des systèmes dynamiques du premier ordre. C'est pourquoi la mémoire d'un événement radiatif passé reste présente dans la température du capteur et décroît exponentiellement avec une échelle de temps correspondant à ce que l'on appelle la constante de temps thermique du capteur. Nous proposons une méthode de traitement d'image qui élimine, au moins partiellement, l'effet mémoire des caméras microbolométriques. Plus précisément, dans toute image, la mémoire des événements radiatifs qui ont eu lieu avant l'image précédente peut être complètement effacée. Le flou de bougé est donc considérablement réduit. Grâce à cette procédure brevetée, on peut considérer que la caméra microbolométrique passe d'un capteur à constante de temps à un capteur à temps d'intégration, comme les détecteurs quantiques refroidis. Ce temps d'intégration spécifique correspond au délai entre deux cycles de lecture. C'est pourquoi il est possible de réduire davantage le flou de bougé en augmentant la fréquence de lecture de la caméra.L'avantage de cette méthode est qu'elle n'implique pas de procédure de déconvolution compliquée et qu'elle est facile à mettre en œuvre et en temps réel. Elle ne dépend pas non plus de la direction du mouvement relatif. Elle considère chaque pixel individuellement et supprime du signal actuel l'influence de l'historique du rayonnement incident qui est plus ancien que la lecture du pixel précédent
Sensibilité à la forme des pales d'un rotor isolé pour l'optimisation aéroacoustique à l'aide de la méthode adjointe discrète
No full textInternational audienceA discrete adjoint framework is developed to optimize rotor self-noise from steady fluid simulations in the rotating frame. To this end, a simplified expression of the off-body frequency-domain Ffowcs-Williams and Hawkings (FW-H) equation is derived for far-field observers, following the model of Hanson and Parzych (1993) originally written for on-body surfaces. The latter is implemented and compared against the results given by an established time-domain FW-H solver. Far-field acoustic pressure sensitivities are derived analytically and validated by comparison with second-order accurate finite differences. The sensitivities of any objective function expressed in terms of the acoustic pressure can therefore be reconstructed. Then the discrete adjoint of a Reynolds-averaged Navier-Stokes solver provides the objective function gradients with respect to the blade shape parameters. The complete workflow is validated against finite difference evaluations on an isolated open rotor in cruise conditions.Un cadre adjoint discret est développé pour optimiser le bruit propre du rotor à partir de simulations de fluides stationnaires dans le domaine tournant. À cette fin, une expression simplifiée de l'équation de Ffowcs-Williams et Hawkings (FW-H) dans le domaine fréquentiel perméable est dérivée pour les observateurs en champ lointain, suivant le modèle de Hanson et Parzych (1993) écrit à l'origine pour les surfaces solides. Ce dernier est mis en œuvre et comparé aux résultats donnés par un solveur FW-H établi dans le domaine temporel. Les sensibilités à la pression acoustique en champ lointain sont dérivées analytiquement et validées par comparaison avec des différences finies précises du second ordre. Les sensibilités de toute fonction objective exprimée en termes de pression acoustique peuvent donc être reconstruites. Ensuite, l'adjoint discret d'un solveur de Navier-Stokes à moyenne de Reynolds fournit les gradients de la fonction objectif par rapport aux paramètres de forme de la pale. L'ensemble de chaîne complète est validé par des évaluations par différences finies sur un rotor ouvert isolé dans des conditions de croisière
Comparaison entre des approches géométrique et de modèle de paroi linéarisé pour les méthodes de frontières immergées
No full textInternational audienceAn in-depth comparison and analysis of the two wall-modeling approaches for RANS of high Reynolds number flows for the immersed boundary method (IBM) is presented. The two wall-modeling approaches are the geometrical approach (GA) proposed by Constant et al. [1] and the linear approach (LA) proposed by Tamaki et al. [2] and correspond to the two most promising methods to obtain oscillation free and accurate results for wall-modeling RANS in an immersed boundary context. To provide an unbiased and accurate comparison, an iso-grid, iso-solver, and iso-numerical method approach is pursued and various 2D and 3D test cases are considered. For 2D, the canonical NACA 0012 and multi-element airfoil are considered whilst the 3D test case considered is an ellipsoid. Given that the GA has been developed for external aerodynamic flows with target applications such as wings and airplanes, it tends to yield more accurate results for these test cases given appropriate user-input parameters. However, the user-input independent approach of the LA provides a more general method that is able to consistently provide accurate results for a large range of flows. Though not as accurate as the GA for airfoils, wings, and airplanes, it tends to yield more accurate results for test cases outside of the target application and best practice guidelines of the GA
Inferring wall-bounded coherent structures from two-dimensional turbulent fields via SHAP analysis
No full textInternational audienceThe analysis of coherent structures in turbulent flows is essential for understanding and controlling turbulence. Explainable Deep Learning (XDL) has been successfully used to identify the most important flow structures in turbulent channels [1], but its computational cost grows significantly as Reynolds number increases. This study adapts the XDL framework to assess whether observations from time-resolved volumetric data of direct numerical simulations (DNS) can be inferred using two-dimensional (2D) slices. Limiting the analysis to 2D subsets reduces the cost of three-dimensional (3D) post-processing and facilitates its application to planar particle image velocimetry (PIV) experimental data. We employ SHapley Additive exPlanations (SHAP), a game-theoretic method, to identify dynamically relevant flow regions by assigning importance scores. Using DNS data from a turbulent channel flow at a friction Reynolds number of 125, we compare SHAP-based structures derived from 3D and 2D data with classically studied structures, including Q events, streaks, and vortex clusters. Our findings show that SHAP-based structures from 2D data statistically agree with those from full 3D fields and previously reported in the literature. These results underscoreth potential of SHAP analyses for studying coherent structures with 2D data, enabling experimental and DNS research at higher Reynolds number
ECRTS Industrial Challenge: Designing a Resilient Time-Aware Shaper Configuration for TSN
No full textInternational audienceIn this paper, we address the ECRTS industrial challenge of reconfiguring at runtime a TSN (Time-Sensitive Networking) network connecting avionics systems, specifically in response to network failures. For this purpose, we propose WPEx-TAS, a novel schedule class that enables resilient and incremental TAS (Time-Aware Scheduling) reconfiguration by allowing critical data flows to be rerouted to already existing alternative time windows in case of failure, without disrupting the pre-computed TAS schedule. By preserving the initial certified schedule, our method reduces the cost and time required for recertification, offering an efficient solution for dynamic network reconfiguration. Furthermore, our approach optimizes bandwidth usage, ensuring that unused bandwidth in TAS windows can be allocated to non-critical traffic. Conducted experiments on the industrial challenge use-case show that a WPEx-TAS schedule can be built in an admissible time, and that the incremental insertion of new streams can be done is less than one second
Inertia-induced power-law scaling in martensites
No full textMartensites subjected to quasistatic deformation are known to exhibit power law distributed acoustic emission in a broad range of scales, however, the origin of the observed scaling behavior and the mechanism of self-organization towards apparent criticality remains obscure. Here we argue that the power law structure of intermittent fluctuations can be at least partially interpreted as an effect of inertia. We build on the insight that inertial dynamics, evidenced by acoustic emission, can become an important factor if the underlying mechanical system is only marginally stable. We first illustrate the possibility of inertia-induced heavy-tailed avalanche size distributions using a prototypical example of a discrete chain with bi-stable springs. We then explore the effects of inertia in fully realistic two-and three-dimensional continuum models of elastic phase transitions. In particular, we demonstrate that a three-dimensional model can produce not only qualitative but also quantitative agreement with experiment.</div
On inertial forces (indirect terms) in problems with a central body
No full textInternational audienceGravitational systems in astrophysics often comprise a body – the primary – that far outweights the others, and which is taken as the centre of the reference frame. A fictitious acceleration, also known as the indirect term, must therefore be added to all other bodies in the system to compensate for the absence of motion of the primary. In this Research Note, we first stress that there is not one indirect term but as many indirect terms as there are bodies in the system that exert a gravitational pull on the primary. For instance, in the case of a protoplanetary disc with two planets, there are three indirect terms: one arising from the whole disc, and one per planet. We also highlight that the direct and indirect gravitational accelerations should be treated in a balanced way: the indirect term from one body should be applied to the other bodies in the system that feel its direct gravitational acceleration, and only to them. We point to situations where one of those terms is usually neglected however, which may lead to spurious results. These ideas are developed here for star-disc-planets interactions, for which we propose a recipe for the force to be applied onto a migrating planet, but they can easily be generalized to other astrophysical systems