Thèse le Speckle Dynamique Polarisé Cadre Théorique et Pratique pour le Traitement du Signal Associé H/F - Doctorat.Gouv.Fr

Établissement : Institut Polytechnique de Paris École polytechnique École doctorale : Ecole Doctorale de l'Institut Polytechnique de Paris Laboratoire de recherche : LPICM - Laboratoire des Interfaces et des Couches Minces Direction de la thèse : Razvigor OSSIKOVSKI ORCID 0000000270847579 Début de la thèse : 2026-10-01 Date limite de candidature : 2026-09-01T23:59:59 Lorsqu'une source de lumière cohérente illumine une surface et qu'une caméra capture le signal rétrodiffusé, on obtient une image de speckle qui résulte de l'interférence destructive ou constructive des différents éléments de diffusion de la surface ou du volume imagé. Le speckle dynamique est une technique d'imagerie du mouvement, car lorsqu'un échantillon se déplace, le speckle associé se déplace également. Les variations temporelles du speckle sont décrites par des statistiques de premier et de second ordre qui sont sensibles à la corrélation entre les champs de speckle imagés à différents moments. Pour obtenir une bonne qualité d'image, il est important d'adapter le protocole d'acquisition des images brutes du champ de speckle à la dynamique du mouvement à capturer. Malgré les travaux intensifs menés dans ce domaine ces dernières années par plusieurs équipes, il n'existe pas de méthode générale permettant d'optimiser la mesure du speckle dynamique associé à une scène en mouvement arbitraire. Un objectif important de cette thèse sera de développer un cadre théorique, basé sur des preuves expérimentales, qui permettra d'obtenir des images de qualité optimale en termes de temps de mesure, de résolution spatiale et de contraste visuel.Par analogie avec le speckle dynamique, la polarimétrie d'imagerie présente un grand potentiel pour la caractérisation des milieux diffusants tels que les tissus ou les plantes. La polarisation de la lumière est sensible à l'organisation de la matière à différentes échelles et peut être utilisée comme facteur d'amélioration du contraste des images de manière non invasive et sans contact. La polarimétrie a été utilisée avec succès dans des applications telles que la discrimination des tissus, la détection précoce de maladies ou la caractérisation de différents stress environnementaux chez les plantes . La polarimétrie utilise une variété d'observables basés sur les propriétés fondamentales de la polarisation de la lumière, en particulier le degré de polarisation, pour révéler des structures cachées et renforcer les contrastes dans les spécimens . Une utilisation assez simple de la polarisation de la lumière utilisant des états de polarisation croisée aux stades de l'illumination et de l'imagerie d'un système d'imagerie de contraste par speckle laser (LSCI) s'est avérée très efficace pour améliorer la qualité de l'image]. Nous allons étudier le lien entre les mécanismes qui déterminent la décorrélation du speckle temporel et le degré de polarisation de la lumière. Dans un article récent, nous avons proposé une première extension de la notion de contraste de speckle temporel au cas polarimétrique. Ainsi, un des objectifs de cette thèse sera de combler ces lacunes en proposant une approche globale pour les signaux de speckle dynamique polarisés. Nous prévoyons de réaliser deux études de cas concrets représentant différents types de mouvement. Le premier concerne la visualisation de la micro circulation sanguine dans la peau humaine et les organes internes. Le second exemple concerne l'imagerie et la quantification du flux de sève dans les feuilles des plantes, ce qui est intéressant pour évaluer leur capacité à échanger des gaz avec l'atmosphère, tels que la vapeur d'eau ou le CO2, ou comme indicateur de la santé des plantes. Ces deux cas d'étude ont été choisis non seulement pour leur impact sociétal potentiel, mais aussi parce qu'ils représentent des conditions physiques très différentes. La circulation du sang et celle de la sève peuvent être considérées comme des mouvements rapides et lents, respectivement, et les propriétés des diffuseurs de lumière dans le sang et la sève sont très différentes l'une de l'autre. La circulation du sang et de la sève sont des références parfaites pour réaliser des expériences liées à des applications réelles afin de démontrer l'intérêt d'un protocole unifié pour réaliser l'imagerie LSCI du speckle dynamique seul ou en combinaison avec la polarimétrie. Detecting and quantifying motion in complex systems is a fundamental challenge in a wide range of applications. In medicine, for example, the ability to sense blood flow can increase surgical precision, improve the study of burns and dermatological reconstruction, facilitate early detection of skin cancer, detect early stage diabetes, and optimize therapeutic interventions. The circulation of sap which takes place in leaves, stems, or roots, has an intrinsic interest in different fields such as physics, engineering, biology, and even climatology. The characterization of the subtle movements occurring in vegetables, including sap circulation, is relevant because they are linked to several essential processes in plant physiology, therefore it can be used as a direct and indirect control to monitor variables intrinsically linked to the biology of plants and their response to their environment such as temperature/humidity stress, gas and water exchange with the atmosphere, response to infection, pollutants or phytosanitary treatments among others. In the fields of biology and ecology, understanding the dynamics of sap flow in plants is critical to understanding growth processes and responses to environmental changes.When we consider human visual perception, it is not just the ability of our eyes to detect motion, but also the ability of our brain to interpret this information in a broader context. This interpretation involves not only detecting changes in position, but also understanding the environment, the lighting conditions, and the relationships between observed objects. Similarly, in computer vision, machines and algorithms perceive and analyze motion through subtle correlations between images. This approach extends beyond the traditional representation of real object space (lengths) to include momentum space, where particle velocities and their distributions are also captured. Momentum space provides a framework for directly measuring dynamic properties that are otherwise invisible in object space. Dynamic speckle exploits the fact that when a coherent wave, such as laser light, interacts with matter, it produces speckle patterns that encode micro-movement and activity in different environments.The speckle phenomenon occurs when coherent light is reflected or scattered by a rough surface or particles in a transparent medium, creating a granular appearance [Goodman 1976]. This pattern, observed on a screen or captured by imaging devices, reveals the dynamic behavior of the scattering particles. The dynamic speckle technique [Rabal 2018] further exploits the fluctuations in these patterns caused by the motion of diffusers within the material. By illuminating a surface with a polarized coherent light source, such as a laser, the backscattered signal is captured and converted into images. By repeating this process over time, we can form a temporal stack of images and transform them into a single activity image using special signal processing techniques [Briers 2013]. This makes dynamic speckle a powerful tool for visualizing subtle motion that is often undetectable by conventional imaging methods.The motion representation of artificial vision depends on the following parameters: i) temporal parameters: integration time, repetition period, and total observation time, ii) light beam parameters: wavelength, polarization of the emitted wave and that of the received wave, and iii) spatial sensor (usually a camera) parameters: pixel size, speckle grain size, noise level, dynamic range, resolution.Despite the many advantages of dynamic speckle techniques, including low cost and non-invasiveness, their widespread application is hampered by the lack of consensus on the hypothetical domains in which motion can be accurately and precisely measured. This lack of agreement results in the absence of general protocols for setting key acquisition parameters and measurement conditions [Okamoto 1993] or signal-to-noise ratio [Zilpelwar 2022], determining optimal wavelengths, and selecting appropriate polarizations for emitted and backscattered waves [Akther 2024]. In addition, there is no unified modeling approach to account for the diversity of potential imaging targets, and the challenge lies in understanding how the 3D structure, symmetry, sample properties [Draijer 2010], and unwanted motion (non-stationary, acceleration...) will affect data processing and interpretation.In its early days, the polarimetric study of speckle fields treated speckle fields as statistical ensembles and described them globally with first- and second-order statistical moments [Li 2002, Zerrad 2010]. More recently, advances in experimental design have enabled polarimetric analysis of speckle fields resolved at the scale of individual speckle grains. These advances have made it possible to experimentally observe the depolarization and repolarization properties [Puget 2012, Zerrad 2013, Staes 2024] on speckle grains, but these physical phenomena are still poorly understood and require further research.Studies coupling dynamic speckle and polarimetry are very recent and scarce because they have been made possible thanks to the latest technical developments in the field of speckle imaging. First, we would like to mention a relevant work that discussed the temporal evolution of the global statistical properties of an objective speckle field [Louie 2025] averaged frame by frame. Authors of this study discussed the optical response of phantom tissues, i.e. samples mimicking the optical properties of real tissues, as well as animal biopsies. Although results are still preliminary, the authors advance the hypothesis of the possibility of using differences in the temporal evolution of speckle fields to discriminate healthy from ill parts of tissues. Moreover, there is the recent development of a portative imager optimized to visualize blood circulation in the skin or organs that may be exposed during a chirurgical operation [Colin 2022]. The innovation of this instrument is that the polarization of the illumination beam is orthogonal to that of the images to minimize the amount of light reflected from the surface of the sample but keeping the signal from light scattered back from the bulk, where the veins are located. The field of speckle polarimetry is at its premises, in the framework of the present Ph.D. we plan to use to investigate the dynamic polarimetric properties of speckle fields at different temporal and spatial scales, i.e. going from the ensemble average of speckles in a whole image or a sequence of images, to that of the individual groups of speckle grains. In particular, we will investigate how sample motion resulting from photothermal effects modify the corresponding speckle polarization distribution. We will investigate the possibility of using the temporal evolution of the speckle statistics as a contrast factor or a way to study the heating dynamics of the sample due to absorption of laser light.The studies on speckle polarization that we will carried within the PhD., will be innovative and of interest in many applications of dynamic speckle imaging. In dynamic scenes that can be described as slow, the speckles show a spatial displacement that is smaller than their own size during the time it takes to capture an image. As a result, the boundaries of individual speckles appear sharp. In analogy to their visual appearance, these speckles are called 'frozen' because they appear to be static in the images. In contrast, under fast motion conditions, the displacement of speckles during the time required to capture an image is comparable to or greater than their own size, and therefore they appear blurred in individual images. Speckles from fast motion scenes are referred to as 'unfrozen' in contrast to the 'frozen' speckles discussed above. At this point, it becomes clear that the terms slow and fast are relative to the characteristic time required to acquire an image, which here plays the role of temporal reference to define the imaging regime. If speckle measurements are sensitive to polarization, sample motion can change the polarization state of the speckles. If such a change is noticeable during the time required to acquire an image, or during the time elapsed between the measurement of successive images, the apparent polarimetric properties of the samples may depend on the acquisition conditions and therefore differ from the intrinsic polarimetric properties of the sample.The main goal of the thesis is to gain a better understanding of the influence of measurement conditions on the temporal statistical properties of dynamic speckle and how they are related to the optical properties of the probed samples. Second, with respect to polarization-sensitive measurements, the thesis also aims to clarify the relation that exists between the characteristic decorrelation time of dynamic speckle fields and their polarimetric properties, in particular their degree of polarization. The achievement of such global goals can be broken down into the following list of partial goals:Objectives related to dynamic speckle:- Model partial correlation between images of frozen speckles: The work will focus on an accurate modeling description of the statistics of fully and partially correlated images of frozen speckles. The work will generalize existing models [Nicolas 2016, Nicolas 2019], obtained during a previous PhD, on the statistical properties of the coefficient of variation in uncorrelated speckle time series. We will enrich this theoretical framework by adding the statistical properties of different figures of merit used in the literature to represent motion, such as the one proposed by Fujii [Fujii].- Model statistical properties of non-frozen speckles: Investigate the laws governing the first and second order statistics of unfrozen speckles. Starting from the ideal case of well-developed speckle fields, we will explore how more realistic intensity distributions affect the values of the statistical parameters describing them.- Study multivariate coefficients of variation in dynamic optical speckle: Considering the effect of the partial correlation that may exist between successive images in a temporal sequence, we will explore different hypothetical regimes of motion, ranging from very fast to extremely slow, and the correlation between successive speckle fields to help understand and implement them in different real-world applications.-Explore statistical properties of non-stationary dynamic speckle models: Many models and methods assume that velocities are constant when estimating motion from optical data, which is not always the case. For example, in the blood circulation, velocity of scatterers in veins is in constant variation due to the influence of heartbeats. To improve current analysis protocols, we will explore new analysis approaches to accurately account for the non-stationary nature (acceleration) of signals.Objectives related to polarimetry - polarimetric speckle- Integrate polarization-sensitive observables (dichroism, retardance, degree of polarization for instance) into statistical models: Address the challenge of incorporating the polarimetric nature of measurements into the statistical analysis framework.- Explore theoretical links between temporal depolarization and characteristic decorrelation time of speckle fields to study how depolarization correlates with temporal contrast phenomena across various dynamic cases. Our methodology combines theoretical modeling and experimental validation in a feedback loop. Experimental evidence serves to inspire theoretical interpretation and to evaluate simulation predictions. On the other hand, simulations provide a fundamental basis for the physical interpretation of experimental results and in many cases inspire new experiments. Using physical models and advanced data processing, one of the aims of the PhD thesis is to explore the possibility of creating a unified model that connects the statistical properties of measured speckles with the physical properties of the proven samples, taking into account the measurement conditions. The thesis will also explore the possibility of combining dynamic speckle with polarimetric imaging to study how the observable parameters of both techniques can be related to the properties of matter.Measuring Dynamic speckle and polarimetric dynamic speckle: unravelling the influence of measurement conditions in motion perceptionThe first goal of this thesis is to explore the connection between the statistical properties of dynamic speckle and the motion characteristics of proven samples and the parameters that determine the measurement of such speckle. The second goal is to deepen our understanding of the relationships between depolarization and motion phenomena in coherent dynamic speckle images. In particular, we aim to refine our models of the temporal decorrelation underlying the dynamic speckle signal and the origins of the depolarization phenomenon.To achieve these goals, we will develop an innovative optical setup to perform dynamic speckle measurements in either polarimetric or non-polarimetric modalities and at optical frequencies (from 400 to 1500 nm) and under well controlled conditions. The optical bench will be capable of performing measurements in forward and back scattering configurations. Since the cadence with which the images are acquired plays a role in the perception of motion and thus in the way the dynamics of the speckle are encoded, we plan to use fast, standard and slow cameras in the setup to be able to study the effect of cadence on the final measurements and also to study the possibility of selecting an optimal cadence given a given rate of change in a sample. The experimental setup will also include the ability to control the polarization of the incident and scattered beams. For this purpose, a removable polarization state generator and a removable polarization state analyzer will be integrated into the illumination and analysis arms, respectively. The combined operation of the polarization state generator and the polarization state analyzer will provide access to an accurate measurement of the polarization phenomena in dynamic speckle through Stokes and Mueller polarimetric imaging modes.This setup will allow the systematic and precise study of several parameters, including: i) the wavelength of the laser source, ii) the characteristics of the diffusers used in the target, iii) the imposed motion characteristics (periodic, aperiodic, velocity, repetition period), iv) the control of the polarization state of the coherent radiation upon emission or reception, and v) the temporal characteristics of the imaging process with respect to the characteristic time of the temporal variations of the imaged object. The various configurations in our experimental simulations will be tailored to the types of objects we wish to image. We plan to study the influence of structural properties of the targets in both polarization and dynamic speckle features: presence of symmetric, oriented, or chiral elements.The use of the optical bench will be closely linked to physical modeling and advanced data processing. The aim is to create a feedback loop where experimental observations suggest refinement of the physical models and vice versa. This iterative approach will allow continuous improvement and validation of both the experimental setup and the models. The integration of these elements will ensure that experimental data and simulations work in synergy, leading to more reliable results across different hypothesis domains.To better understand the statistical properties of dynamic speckles, we propose the following sub-tasks:a. Examine multivariate coefficients of variation in dynamic optical speckle: exploring different hypothetical regimes to enhance understanding and application.b. Measurement of samples under non-stationary motion: measure the impact of acceleration (i.e. velocity variation) during acquisition time in speckle statistics. c. Optimize polarization choices in technology: particularly focusing on scenarios involving motions with high and slow velocities.d. Explore theoretical links between temporal depolarization and decorrelation to establish how depolarization correlates with temporal contrast phenomena across various scenarios.e. Develop a hybrid imaging device that integrates Mueller imaging and dynamic speckle imaging, aiming to leverage the strengths of both techniques for improved imaging capabilities.Speckle statistics: towards a global modelling of polarization sensitive dynamic speckle data at different motion regimesAs mentioned above, the experimental approach is complemented by a rigorous formal methodology where observations are systematized and theoretical models are proposed wherever feasible. These formal tools, which include matrix algebra for describing polarimetric and temporal data, signal processing tools, and statistical models, are being developed in an ongoing collaboration among the thesis supervisors.Speckle statistics, extensively studied for static speckle, is an important physical phenomenon related to wave interference. When a wave is projected onto a surface, its backscatter results from the coherent addition of signals emitted by various randomly distributed elements within the resolution cell. This sum of waves, depending on their relative phases, produces constructive and destructive interferences that form the speckle. According to Godmann [Goodman 1976], when the speckle results from the sum of independent contributions from N complex response elements, a single parameter, sigma, can be used, which is related to both the standard deviation of the real and imaginary parts of the complex signal. As a result, the electromagnetic field amplitude and intensity can be described by the Rayleigh and exponential distributions, respectively. Summing multiple independent speckles results in a speckle that is no longer fully developed and requires an additional parameter to account for the number of terms in the sum. In this case, the amplitude follows a Rayleigh-Nakagami distribution defined by two parameters: scale and shape.When we encounter a speckle pattern that is not fully developed, or when we analyze mixtures of statistical populations with different parameters, the resulting distributions change. Similarly, the statistical laws of dynamic speckle, which vary with time, are very different from those of static speckle and have not been extensively studied [Yoshimura 1986]. Our previous research has revealed several hypothetical frameworks for the analysis of dynamic speckle, including: i) a stationary regime versus a non-stationary regime, ii) a frozen speckle zone where speckle decorrelation is minimal during the integration time, but some decorrelation occurs before the end of the sequence due to detectable motion with images taken at sufficiently long time intervals, and, iii) a zone of completely uncorrelated speckle patterns.To better understand the statistical properties of dynamic speckles we propose to following partial tasks: a. Understand statistical properties of non-stationary models: Develop analysis methods that accurately account for the non-stationary nature of signalsb. Model behavior of partial correlations in frozen speckle: Focus on modeling the behavior of partial correlations within a series of frozen speckle to better understand their dynamic properties and implications.c. Examine laws of unfrozen speckle: Investigate the laws governing unfrozen speckle, noting that the intensity distributions of dynamic speckle do not always align with classical gamma law distributions.d. Explore statistical behaviors of parameters: Expand upon the initial studies by Jean-Marie Nicolas [Nicolas 2019] on the statistical properties of the coefficient of variation in decorrelated speckle time series. Enrich this framework by incorporating the statistical properties of other temporal parameters like those proposed by Fujii and measures of autocorrelation.e. Integrate polarimetric dimensions into statistical models: Address the challenge of incorporating the polarimetric nature of measurements into the statistical analysis frameworkBlood microcirculation: Evaluate blood flow from partially uncorrelated series of no-frozen speckle images with parasitic non-stationary movementsSince its initial development in the 1960s, the dynamic speckle technique has been used to develop numerous medical imaging techniques to better visualize and characterize blood flow in various organs, as well as to quantify flow velocity [Rabal 2009]. In general, we use the speckle contrast, which is a simple ratio between the standard deviation and the mean evaluated over a population of speckle intensities. Various modeling approaches allow to link the contrast to the autocorrelation of the intensity, but suffer from the following drawbacks: i) they are only applicable when the velocities are high enough compared to the acquisition / integration time, ii) they neglect the static contribution of the signal and do not account for multiple scattering to link velocities and decorrelation.Regarding the last point, one of our major contributions was the systematic use of a cross-polarization acquisition configuration to maximize the contribution of multiple scattering over first-order interactions. This allowed us to improve the penetration depths achieved. Another challenge, obtaining an effective velocity parameter from the previous relationship to go beyond qualitative images, is not a trivial problem. We have recently proposed a method that could be classified as semi-quantitative: instead of deriving a velocity or flow parameter, we propose to measure an index that is reproducible over time or between two independent devices. However, as mentioned above, these models are only valid if the signals are uncorrelated from each other and for signals containing a sufficient number of decorrelations.The latest methods, known as quantitative methods, abandon the idea of quantifying from speckle contrast. They rely on high-speed acquisitions to track the speckle grain, assuming there is a correlation between contiguous images. Essentially, this means directly measuring the decorrelation time between frozen and correlated speckle images, implying we are no longer in the previous framework of un correlated speckles. Increasing the acquisition rate can not only provide more direct quantification but also reduce the required acquisition times, particularly for tracking slower movements. This is the context in which we studied the use of ultra-fast acquisitions in Simon Erdmann's thesis [Erdmann 2022]. This thesis illustrates the need to adapt the choice of signal parameters to the hypothesis regimes governing the involved times: integration time, repetition time, and decorrelation time.Parasitic non-stationary motion, such as the heart beating during acquisition, poses an additional challenge to accurate blood flow estimation. To address the challenge of imaging the microvasculature of a perfused heart using dynamic speckle technology without interrupting its natural motion, in the framework of the thesis we will explore the improvement of specific data treatments that allow the selection of images corresponding to very similar positions of the organ using optical flow estimation methods [Plyer 2023].Although in recent years we have explored some original ideas to overcome the main difficulties that hinder blood flow estimation, many challenges remain, such as accurately accounting for the effect of multiple scattered photons, handling partial series of partially correlated speckle images, and quantifying micromovements in the presence of unwanted macromovements. These challenges guide our research goals.a. Enhancement of movement management in medical imaging: Refine existing motion management strategies within calibration procedures to enhance cardiology applications. In particular, explore how polarimetric information can optimize the trade-off between acquisition time and motion-related depolarization data.b. Enhancement of velocity inversion accuracy: Deepen the understanding of the velocity ranges for which velocity inversion remains robust. Clarify the operational hypothesis domain-whether for frozen or decorrelated speckles-and identify the optimal parameters for accurate velocity inversion.c. Advancement of 3D imaging techniques: Development and implementation of advanced 3D imaging techniques for improved visualization and analysis of dynamic motion to enhance diagnostic accuracy and treatment planning in medical applications.Sap microcirculation: Partially correlated series of frozen speckle imagesSap flow can be described as slow motion compared to blood flow. Traditional methods used to visualize blood flow fail when applied to sap circulation and need to be rethought and redefined to take into account the specificities of fluid movement in plants. In particular, and in contrast to blood flow in animals or humans, there are processes in vegetables such as water evaporation and organelle diffusion inside cells (especially that of chloroplasts during the day) that are characterized by a rate of change comparable to that of sap flow. In addition to sap flow, all of these processes simultaneously influence the dynamics of speckle and have to be taken into account during data acquisition, data processing, and physical interpretation of the images to avoid biases and errors compared to the case of blood flow imaging. Moreover, the directionality of such movements, which are characterized by a certain degree of randomness, differs from the deterministic fate of sap flow in the veins. Blood and sap flow appear to be very different when recorded using the same protocol [Colin 2024]. In most of the cases reported in the literature, consecutive frames in a video sequence corresponding to blood flow are temporarily uncorrelated. TThe contrast parameter can be calculated temporally or spatially. It reveals the motion associated with speckle decorrelation. On plant leaves, this temporal decorrelation appears much slower, and successive images appear highly correlated with each other. Therefore, it is clear that a better understanding of the various dynamic phenomena occurring in a leaf is needed to adapt the measurement parameters and the data processing itself. The use of a reduced image acquisition rate to image sap flow is not a practical option, because the time required to measure the data needed for statistics would be prohibitively long for the development of in situ, in-operando plant characterization techniques in a reasonable time frame. For this reason, we decided to use the theoretical model developed in the Ph.D. thesis to optimize the acquisition parameters or to explore alternative options.A careful review of the existing literature shows that a number of algorithms have been developed in recent years to process raw data and improve the image quality of sap circulation in leaves. The pioneering work of [Matsuo 2006] proposed an index defined as the sum of the absolute values of the differences between each individual image and the average image. [Zhong 2016] introduced an innovative approach based on normal vector analysis instead of light intensities. This method conceptualizes the speckle pattern as a triangulated surface, providing a new perspective for analyzing complex speckle patterns by incorporating spatial gradients. [Pieczywek 2017, Pieczywek 2018] proposed a modified version of the conventional Fujii index and illustrated its use to image the development of a fungal infection in mature apples. The wavelength of the light used for illumination seems to be an important factor to improve the sensitivity of sap flow estimation [Riahi 2006, Shintani 2009].In order to achieve a better imaging and quantification of sap flow in plants we propose the following partial steps:a. Master the hypothesis domain of partially uncorrelated frozen speckles: Enhance understanding and control of the hypothesis domain involving partially uncorrelated frozen speckles, incorporating polarimetric dimensions to advance the technology and refine the selection of indices for use.b. Develop alternative quantification methods to optical flow: Explore and establish alternative methods for quantification that can serve as viable replacements or complements to optical flow techniques.c. Co-design dedicated sensors: Collaborate on the design and development of dedicated sensors tailored to specific needs, improving the effectiveness and integration of new technologies in practical applications.

Le candidat devra être en possession d'un diplôme d'ingénieur délivré par une Grande école, ou bien d'une licence universitaire en sciences assorti d'un diplôme de master en recherche. Le candidat doit avoir un goût appuyé pour les sujets transdisciplinaires impliquant des aspects théoriques (simulation, formulation...) et appliqués (montage de prototypes, bancs d'essai, conception d'expériences). Le sujet de thèse implique un travail informatique tourné vers le traitement automatique d'images par ordinateur et la conception d'expériences impliquant la réalisation automatique de mesures. Le candidat devra donc faire preuve d'une expérience avec un ou plusieurs de ces languages ou outils: MATLAB, LABVIEW, PHYTON (gdal, geopandas, pytorch ...) ou C++,et avoir un traitement du signal et des images. Au cours de la thèse, le candidat devra consulter la bibliographie existante (articles, thèses, communications...) sur le sujet souvent produite par auteurs étrangers, à participer à la rédaction d'articles et à donner des conférences devant d'un public international. Il est donc nécessaire que le candidat maitrise l'anglais à un niveau technique.