Nos centros urbanos, as precipitações sempre estiveram ligadas a problemas como inundações e propagação de doenças. Para solucioná-los, é comum a realização de obras hidráulicas nos sistemas de drenagem urbanos. Para tanto, deve-se conhecer as condições da bacia hidrográfica e as consequências que qualquer alteração no ambiente pode causar. Portanto, modelos hidrológicos são utilizados na previsão do comportamento das bacias frente a eventos de precipitação, aumentando a eficácia das obras e diminuindo os riscos associados a estas. Para o uso de modelos, são necessários diversos parâmetros hidrológicos referentes à bacia, tais como área de drenagem, comprimento e declividade dos talvegues, tipo de cobertura de solo etc. Com o avanço da tecnologia, a determinação destes torna-se cada vez mais precisa, bem como os modelos utilizados, trazendo o Sistema de Informações Geográficas (SIG) e o sensoriamento remoto como poderosas ferramentas de apoio a estudos hidrológicos. Neste trabalho, aplicou-se o processo de classificação automática supervisionada pelo método da Análise Orientada a Objeto a uma imagem de satélite de alta resolução da bacia hidrográfica do córrego do Gregório, para caracterizar sua cobertura de solo e determinar os parâmetros hidrológicos número de deflúvio (CN, pelo método do SCS), grau de vegetação (PP), área (A), comprimento (L) e declividades dos talvegues (S) das sub-bacias que compõem a bacia, para as quais os resultados obtidos foram bastante satisfatórios. Por fim, atualizou-se o modelo hidrológico EESC (1993), referente ao sistema de macrodrenagem de São Carlos, obtendo-se hidrogramas finais com diferenças, em relação ao modelo original, de até 33,96% para vazão de pico (Qp), 77,78% para tempo de pico (tp) e 29,86% para volume total de escoamento. / In urban centers, precipitation always been related to problems such as floods and spread of disease. To solve them, it is common to make hydraulic interventions in the urban drainage systems. For this, it is necessary to know the conditions of the watershed and the consequences that any change in the environment can cause. Therefore, hydrological models are used to predict the river behavior in opposite to precipitation events, increasing the efficiency of the hydraulic interventions and reducing the associated risks to these. For the use of models, it is necessary to have several hydrological parameters related to the basin such as drainage area, river length, slope of the thalweg, type of soil cover etc. Trough the technological advancement, the parameter determination becomes more accurate as well as the models, and the Geographic Information System (GIS) and remote sensing appear as powerful tools to support hydrological studies. In this study, we have applied the automatic supervised classification process by the Object-Oriented Analysis method to a high resolution satellite image of the córrego do Gregório watershed, to classify soil coverage and to determine the hydrological parameters curve-number (CN by the SCS method), vegetation degree (PP), area (A), length (L), and slope of thalwegs (S) of the sub-basins of the córrego do Gregório watershed, for which the results were quite satisfactory. Finally, a hydrological model for the São Carlos macrodrainage system called EESC model (1993) was updated with the new parameters, obtaining final hydrographs with differences from the original model up to 33.96% for peak discharge (Qp), 77.78% for peak time (tp) and 29.86% for total volume of runoff.

Les nouveaux capteurs satellitaires permettent l'acquisition d'images d'un très haut niveau de détail à des cadences élevées, produisant ainsi une importante masse de données. Le traitement manuel de ces données étant devenu impossible, de nouveaux outils sont nécessaires afin de les traiter automatiquement. Des algorithmes de segmentation efficaces sont nécessaires pour extraire des objets d'intérêt de ces images. Cependant les segments produits ne correspondent pas aux objets d'intérêt, rendant difficile l'utilisation de connaissances expertes.Dans le cadre de cette thèse nous proposons de changer le niveau d'interprétation d'une image afin de voir les objets d'intérêt pour l'expert comme des objets composés par des segments. Pour cela, nous avons mis en place un processus d'apprentissage multi-niveaux dans le but d'apprendre ces règles de composition. Une règle de composition ainsi apprise peut ensuite être utilisée pour extraire les objets d'intérêt correspondant. Dans un second temps, nous proposons d'utiliser l'algorithme d'apprentissage de règles de composition comme première étape d'une approche montante-descendante. Cette chaîne de traitement a pour objectif d'améliorer la classification à partir des informations contextuelles et de connaissances expertes. Des objets composés de plus haut niveau sémantique sont extraits à partir de règles apprises ou fournies par l'expert, et cette nouvelle information est utilisée pour mettre à jour la classification des objets aux niveaux inférieurs. L'ensemble de ces travaux ont été testés et validés sur des images Pléiades représentant la ville de Strasbourg. Les résultats obtenus montrent l'efficacité de l'apprentissage de règles de composition pour faire le lien entre connaissance experte et segmentation, ainsi que l'intérêt de l'utilisation d'informations contextuelles dans l'analyse d'images de télédétection à très haute résolution spatiale. / The new remote sensors allow the acquisition of very high spatial resolution images at high speeds, thus producing alarge volume of data. Manual processing of these data has become impossible, new tools are needed to process them automatically. Effective segmentation algorithms are required to extract objects of interest of these images. However, the produced segments do not match to objects of interest, making it difficult to use expert knowledge.In this thesis we propose to change the level of interpretation of an image in order to see the objects of interest of the expert as objects composed of segments. For this purpose, we have implemented a multi-level learning process in order to learn composition rules. Such a composition rule can then be used to extract corresponding objects of interest.In a second step, we propose to use the composition rules learning algorithm as a first step of a bottom-up top-down approach. This processing chain aims at improving the classification from contextual knowledge and expert information.Composed objects of higher semantic level are extracted from learned rules or rules provided by the expert, and this new information is used to update the classification of objects at lower levels.The proposed method has been tested and validated on Pléiades images representing the city of Strasbourg. The results show the effectiveness of the composition rules learning algorithm to make the link between expert knowledge and segmentation, as well as the interest of the use of contextual information in the analysis of remotely sensed very high spatial resolution images.

No description available.

Les images à très haute résolution spatiale (THR) telles que les images Pléiades (50 cm en Panchromatique, 2m en multispectral) rendent possible une description fine de la structure forestière (distribution et dimensions des arbres) à l'échelle du peuplement, en exploitant la relation entre la structure spatiale des arbres et la texture d'image quand la taille du pixel est inférieure à la dimension des arbres. Cette attente répond au besoin d'inventaire spatialisé de la ressource forestière à l'échelle du peuplement et de ses changements dus à la gestion forestière, à l'aménagement du territoire ou aux événements catastrophiques. L'objectif est double: (1) évaluer le potentiel de la texture d'images THR pour estimer les principales variables de structure forestière (diamètre des couronnes, diamètre du tronc, hauteur, densité ou espacement des arbres) à l'échelle du peuplement; (2) sur ces bases, classer les données image, au niveau pixel, par types de structure forestière afin de produire l'information spatialisée la plus fine possible. Les principaux développements portent sur l'automatisation du paramètrage, la sélection de variables, la modélisation par régression multivariable et une approche de classification par classifieurs d'ensemble (Forêts Aléatoires ou Random Forests). Ils sont testés et évalués sur deux sites de la forêt landaise de pin maritime à partir de trois images Pléiades et une Quickbird, acquises dans diverses conditions (saison, position du soleil, angles de visée). La méthodologie proposée est générique. La robustesse aux conditions d'acquisition des images est évaluée. Les résultats montrent que des variations fines de texture caractéristiques de celles de la structure forestière sont bien identifiables. Les performances en terme d'estimation des variables forestières (RMSE) : ~1.1 m pour le diamètre des couronnes, ~3 m pour la hauteur des arbres ou encore ~0.9 m pour leur espacement, ainsi qu'en cartographie des structures forestières (~82 % de taux de bonne classification pour la reconnaissance des 5 classes principales de la structure forestière) sont satisfaisantes d'un point de vue opérationnel. L'application à des images multi-annuelles permettra d'évaluer leur capacité à détecter et cartographier des changements tels que coupe forestière, mitage urbain ou encore dégâts de tempête. / Very High spatial Resolution (VHR) images like Pléiades imagery (50 cm panchromatic, 2m multispectral) allows a detailed description of forest structure (tree distribution and size) at stand level, by exploiting the spatial relationship between tree structure and image texture when the pixel size is smaller than tree dimensions. This information meets the expected strong need for spatial inventory of forest resources at the stand level and its changes due to forest management, land use or catastrophic events. The aim is twofold : (1) assess the VHR satellite images potential to estimate the main variables of forest structure from the image texture: crown diameter, stem diameter, height, density or tree spacing, (2) on these bases, a pixel-based image classification of forest structure is processed in order to produce the finest possible spatial information. The main developments concern parameter optimization, variable selection, multivariate regression modelling and ensemble-based classification (Random Forests). They are tested and evaluated on the Landes maritime pine forest with three Pléiades images and a Quickbird image acquired under different conditions (season, sun angle, view angle). The method is generic. The robustness of the proposed method to image acquisition parameters is evaluated. Results show that fine variations of texture characteristics related to those of forest structure are clearly identifiable. Performances in terms of forest variable estimation (RMSE): ~1,1m for crown diameter, ~3m for tree height and ~0,9m for tree spacing, as well as forest structure mapping (~82% Overall accuracy for the classification of the five main forest structure classes) are satisfactory from an operational perspective. Their application to multi- annual images will assess their ability to detect and map forest changes such as clear cut, urban sprawl or storm damages.

Dans un contexte de changements globaux, il est primordial de mettre au point des pratiques de gestion durable des forêts tropicales assurant le maintien de services environnementaux clés (e.g., biodiversité, stockage de carbone) et la production de biens essentiels aux communautés locales. La mesure, la spatialisation et la compréhension des déterminismes de la structure des forêts tropicales est un challenge pour la gestion durable des ressources forestières. Les forêts denses humides (FDH) de Nouvelle-Calédonie, un point chaud de biodiversité localisé dans le Pacifique Sud-Ouest, sont un modèle d'étude idéal pour comprendre les déterminants de la structure des FDH. En effet, les FDH néo-calédoniennes sont réparties le long d’une chaîne de montagne et sont ainsi soumises à de forts gradients environnementaux auxquels se superposent différents gradients de perturbations naturelles ou anthropiques. Les objectifs de cette thèse sont (i) de définir les caractéristiques structurales des FDH néo-calédoniennes, (ii) de cartographier les FDH et prédire leur structure à large échelle, et (iii) de quantifier l'influence de l’environnement et des dynamiques forestières sur la structure des FDH. Pour cela, 23 parcelles d’inventaire forestier de 100 m x 100 m ont été mise en place en Province Nord, entre 250 et 900 m d'altitude et 1500 et 3000 mm de précipitations annuelles. En plus de caractériser localement la structure des FDH, ces parcelles ont permis de calibrer un modèle prédictif basé sur l’analyse de la texture de la canopée, à l'aide la méthode FOTO (FOurier transform Textural Ordination), qui a été appliqué à huit images satellitaires à très haute résolution Pléiades (couvrant 1295 km2). Un tel modèle capable de lier texture et structure repose sur le postulat que la relation allométrique entre le DBH (Diameter at Breast Height) et l'aire de la couronne des arbres de canopée est stable. Nous avons également testé cette relation à échelle pantropicale. Nos résultats ont montré que les FDH néo-calédoniennes sont denses (1182 ± 233 tiges/ha), ont une aire basale élevée (44 ± 11 m2/ha), une canopée relativement basse (14 ± 3 m) et une biomasse aérienne caractéristique des forêts tropicales (299 ± 83 t/ha). Elles se distinguent également par une importante variabilité structurale. Cette variabilité est du même ordre que ce soit le long de gradients environnementaux ou de gradients de succession forestière. La méthode FOTO appliquée aux images Pléiades a permis de prédire et de spatialiser des paramètres structuraux clefs (tels que la densité de tiges et la biomasse aérienne des FDH) à partir de corrélations robustes avec les indices de texture de la canopée (R² ≥ 0,6; RMSE ≤ 20%). La structure des FDH est principalement dirigée par l'insolation potentielle et l'altitude à l'échelle des massifs montagneux, et par la pente et un indicateur topographique d'humidité à l'échelle du versant. Ces travaux permettront d'estimer les ressources forestières à l'échelle de la Nouvelle-Calédonie et de définir une nouvelle typologie des FDH sur le territoire intégrant leur variabilité structurale. / In the course of global change, new practices of sustainable management in tropical rainforests that maintain key environmental services (e.g., biodiversity, carbon sequestration) and produce goods on which local communities rely is needed. The measurement, spatialization and understanding of the drivers of rainforest structure at large scale is challenging for managing sustainably forest resources. Rainforests of New Caledonia, a biodiversity hotspot located in the South-West Pacific, are a well-suited study model to explore the drivers of rainforest structure. Indeed, New Caledonian rainforests are distributed along a mountain chain, which creates strong environmental gradients overlaid by a range of natural and anthropogenic disturbance gradients. The aims of this thesis are (i) to define some structural features of New Caledonian rainforests, (ii) to map rainforests and to predict their structure at large scale, and (iii) to quantify the influence of the environment and the forest dynamics on rainforest structure. To this end, 23 one hectare forest inventories were set up in the North Province of New Caledonia. In these plots, elevation ranged between 250 and 900 m and annual rainfall between 1500 and 3000 mm. In addition to characterize locally rainforest structure, these plots were used to calibrate a predictive model based on a textural analysis of the canopy, using the FOTO (FOurier transform Textural Ordination) method, which was applied to eight very high resolution images from a Pléiades satellite (covering 1295 km2). Such a model able to relate texture and structure is based on the hypothesis that the allometric relationship between the DBH (Diameter at Breast Height) and the crown size of a canopy tree is stable. We tested this hypothesis tropics-wide. Our results show that New Caledonian rainforests are dense (1182 ± 233 tree/ha), with a high basal area (44 ± 11 m2/ha), a relatively low canopy (14 ± 3 m) and an above-ground biomass typical of tropical rainforests (299 ± 83 t/ha). These forests are also characterized by a high structural variability. This variability has the same range when influenced by environmental gradients as when influenced by forest succession gradients. The FOTO method applied to Pléiades images allowed to predict and spatialize key structural parameters (like the stem density or the above-ground biomass of rainforests) from robust correlations with the textural indices of the canopy (R² ≥ 0,6; RMSE ≤ 20%). The structure of New Caledonian rainforest is mainly driven by the potential insolation and the elevation at the scale of mountain massifs, and by the slope and the topographic wetness at the scale of a mountainside. These findings will enable to estimate rainforest resources across the territory and to define a new typology of New Caledonian rainforests taking into account their structural variability.

The present work introduces a new method that is used to explore the connection between molecular order and molecular dynamics in liquid crystals. In liquid crystals, the building blocks show a liquid like disorder in at least one dimension of space with an otherwise crystalline like positional or orientational long range order. A new microscope is introduced that combines polarization measurements with the ability to track single fluorescent probe molecules in a thin sample of ordered liquid crystal. A new method for the analysis of orientation dependent diffusion is also introduced. It can be used to spatially resolve the anisotropic diffusion of the probe molecules. With this setup, molecular structure and molecular dynamics can be directly compared on a μm scale. Three different kinds of liquid crystal samples are analyzed with the new experimental method. First, twisted nematic liquid crystal cells are used to verify a proposed model for the connection between molecular structure and the dynamics in twisted nematic cells. Second, the liquid crystal structure and probe mobility are analyzed in homogeneous samples in a temperature regulated environment. The third experiment focuses on the combination of both of these scenarios. Different domains in a heterogeneous section of a sample are analyzed with different methodical approaches at various temperatures. The results display the close connection between molecular order and molecular dynamics in the samples. It is also found that the probe molecules introduce local distortions in the director field of the host material. Despite this realization, only the absolute value of the probes mobility seems to be effected. The anisotropy of the translational diffusion of the probe molecules resembles the results found in the literature on the self-diffusion of the liquid crystal molecules. The anisotropy also follows the same temperature dependence as the order of the host molecules. Using these results and the new method of analyzing single molecule tracking data, it is shown that the structure of a heterogeneous sample can be spatially resolved, only by means of single probe molecule tracking.

Recent interests in computer based tools and simulations for PET imaging studies have been a leading source for many new developments. A strong emphasis in these studies has been to improve and optimize the PET scanners for better image quality and quantification of related system parameters. In this project, an attempt has been made to develop a Matlab tool intended to be of educational nature for new students where one can perform demonstration of PET-like imaging in a simple and quick way. This demonstration tool utilizes a high resolution, voxel based digital brain (Zubal) phantom as a primary study object. A tumor of specific size is defined by the user on a chosen slice of the phantom. The output images from this tool show the exact location of the predefined tumor. The algorithm attempts to estimate the positron emission direction, positron range distribution and photon detection in a circular geometry. Additional attempt has been made to estimate certain statistical parameters against a specific amount of radiotracer uptake. These include spatial resolution, photons count, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the ultimate PET image. Dependence of these estimated results by the tool on different system input parameters has been studied.

No description available.

Semiconductor detectors are attracting interest for use in photon-counting spectral computed tomography. In order to obtain a high spatial resolution, it is of interest to find the photon interaction position. In this work we investigate if machine learning can be used to obtain a sub-pixel spatial resolution in a photon-counting silicon strip detector with pixels of 10 µm. Simulated charge distributions from events in one, three, and seven positions in each of three pixels were investigated using the MATLAB® Classification Learner application to determine the correct interaction position. Different machine learning models were trained and tested in order to maximize performance. With pulses originating from one and seven positions within each pixel, the model was able to find the originating pixel with an accuracy of 100% and 88.9% respectively. Further, the correct position within a pixel was found with an accuracy of 54.0% and 29.4% using three and seven positions per pixel respectively. These results show the possibility of improving the spatial resolution with machine learning. / Halvledardetektorer är av stigande intresse inom forskning för användning i fotonräknande datortomografi med spektral upplösning. För att erhålla en hög spatiell upplösning är det av intresse att hitta fotonens ursprungliga interaktionsposition. I detta arbete undersöks om maskininlärning kan användas för att erhålla en spatiell upplösning på subpixelnivå i en fotonräknande kiselstrippdetektor med 10 µm pixlar. Laddningsfördelningen från simulerade interaktioner i en, tre, och sju positioner inom var och en av tre pixlar undersöktes med hjälp av applikationen Classification Learner i MATLAB® för att bestämma den korrekta interaktionspositionen. Olika maskininlärningsmodeller tränades och testades för att maximera prestandan. När pulser från en och sju positioner inom pixeln användes, kunde modellen hitta den korrekta pixeln med en noggrannhet på 100% respektive 88.9%. Vidare kunde den korrekta positionen inom en pixel bestämmas med en noggrannhet på 54.0% och 29.4% när tre respektive sju positioner inom varje pixel användes. Resultaten visar att det skulle vara möjligt att förbättra den spatiella upplösningen med hjälp av maskininlärning.

No description available.