Étant donnée l'évolution technologique des capteurs embarqués à bord des satellites, le potentiel d'images satellitaires accessible s'accroît de telle manière que se pose maintenant la question de son exploitation la plus efficace possible. C'est l'objectif du projet CARTAM-SAT que de fluidifier la chaîne de traitement depuis les satellites jusqu'aux utilisateurs des images. La thèse s'inscrit dans ce cadre. Les traitements relatifs aux images ont évolué au cours des années. Les images basse résolution étaient traitées par une approche dite pixel alors que la haute résolution a permis le développement d'une approche dite objet. Cette dernière s'attache à analyser non plus des pixels isolés, mais des groupes de pixels représentatifs d'objets concrets sur le terrain. Ainsi, en principe, ces groupes de pixels sont dotés d'une sémantique propre au domaine de la télédétection. La représentation des connaissances a évolué parallèlement aux images satellitaires. Les standards de représentation ont profité de l'expansion du web pour donner naissance à des standards comme OWL. Celui-ci repose en grande partie sur les logiques de description qui permettent l'utilisation de raisonneurs automatiques capables d'inférer une connaissance implicite.Cette thèse se place à la jonction de ces deux sciences et propose une approche ontologique d'analyse des images satellitaires. Il s'agit de formaliser différents types de connaissances et de conceptualisations implicitement utilisés par les logiciels de traitement d'image et par les experts en télédétection, puis de raisonner automatiquement sur la description d'une image pour en obtenir une interprétation sémantique.Ce principe général est susceptible de nombreuses déclinaisons techniques. La mise en œuvre a consisté en la réalisation d'un prototype alliant une bibliothèque d'analyse d'images satellitaires et un raisonneur basé sur les ontologies. L'implémentation proposée dans la thèse permet d'explorer quatre déclinaisons techniques du principe qui mènent à des discussions sur la complémentarité des paradigmes d'analyse pixel et objet, la représentation de certaines relations spatiales et la place de la connaissance par rapport aux traitements. / Given the technological development of embedded satellite sensors, the potential of available satellite images increases so that the question now arises of their most efficient exploitation possible. This is the purpose of the CARTAM-SAT project to fluidize the processing workflow from satellite images to users. The thesis is part of this framework.Processing operations relating to images have evolved over the years. Low-resolution images were processed by a so-called pixel approach while the high-resolution has allowed the development of a so-called object approach. The latter focuses on analysing not about the isolated pixels, but about groups of pixels representative of concrete objects on the ground. Thus, in principle, these are groups of pixels with a domain-specific remote sensing semantics.Along with satellite imagery, knowledge representation has evolved. The standards of representation have benefited from the expansion of the web to give rise to standards like OWL. This one is widely based on description logics that allow the use of automated reasoners able to infer implicit knowledge.This thesis is at the junction of these two sciences and provides an ontological approach for analysing satellite images. The aim is to formalize different types of knowledges and conceptualizations implicitly used by image processing programs and by remote sensing experts, and then reasoning automatically on an image description to obtain one semantic interpretation.This general principle may have numerous technical variations. The implementation consisted in a prototype combining a satellite image analysis library and an ontology-based reasoner. The implementation proposed in the thesis allows to explore four technical variations of the principle that lead to discussions on the complementarity of pixel and object analysis paradigms, the representation of some of the spatial relations and the role of knowledge in relation to processing.

The human vision system is generally recognized as being superior to all known artificial vision systems. Visual attention, among many processes that are related to human vision, is responsible for identifying relevant regions in a scene for further processing. In most cases, analyzing an entire scene is unnecessary and inevitably time consuming. Hence considering visual attention might be advantageous. A subfield of computer vision where this particular functionality is computationally emulated has been shown to retain high potential in solving real world vision problems effectively. In this monograph, elements of visual attention are explored and algorithms are proposed that exploit such elements in order to enhance image understanding capabilities. Satellite images are given special attention due to their practical relevance, inherent complexity in terms of image contents, and their resolution. Processing such large-size images using visual attention can be very helpful since one can first identify relevant regions and deploy further detailed analysis in those regions only. Bottom-up features, which are directly derived from the scene contents, are at the core of visual attention and help identify salient image regions. In the literature, the use of intensity, orientation and color as dominant features to compute bottom-up attention is ubiquitous. The effects of incorporating an entropy feature on top of the above mentioned ones are also studied. This investigation demonstrates that such integration makes visual attention more sensitive to fine details and hence retains the potential to be exploited in a suitable context. One interesting application of bottom-up attention, which is also examined in this work, is that of image segmentation. Since low salient regions generally correspond to homogenously textured regions in the input image; a model can therefore be learned from a homogenous region and used to group similar textures existing in other image regions. Experimentation demonstrates that the proposed method produces realistic segmentation on satellite images. Top-down attention, on the other hand, is influenced by the observer’s current states such as knowledge, goal, and expectation. It can be exploited to locate target objects depending on various features, and increases search or recognition efficiency by concentrating on the relevant image regions only. This technique is very helpful in processing large images such as satellite images. A novel algorithm for computing top-down attention is proposed which is able to learn and quantify important bottom-up features from a set of training images and enhances such features in a test image in order to localize objects having similar features. An object recognition technique is then deployed that extracts potential target objects from the computed top-down attention map and attempts to recognize them. An object descriptor is formed based on physical appearance and uses both texture and shape information. This combination is shown to be especially useful in the object recognition phase. The proposed texture descriptor is based on Legendre moments computed on local binary patterns, while shape is described using Hu moment invariants. Several tools and techniques such as different types of moments of functions, and combinations of different measures have been applied for the purpose of experimentations. The developed algorithms are generalized, efficient and effective, and have the potential to be deployed for real world problems. A dedicated software testing platform has been designed to facilitate the manipulation of satellite images and support a modular and flexible implementation of computational methods, including various components of visual attention models.

De virvelgator som bildas i atmosfären bakom höga berg på öar påminner till utseendet starkt om de periodiska flöden som uppstår vid strömning kring en cirkulär cylinder. Friktionen mellan fluiden och cylinderns yta gör att det bildas en vak nedströms cylindern. Periodisk virvelspridning där von Kármán-virvlar sänds ut kan uppstå. Utifrån Reynolds tal går det att karaktärisera strömningen och med hjälp av en numerisk modell kan tillstånden simuleras. Saknas en turbulensmodell i algoritmen blir resultaten för höga Reynolds tal felaktiga. De atmosfäriska virvelgatorna uppstår dock inte på grund av friktion. Istället krävs blockering av luftmassor och variationer i densitet för att virvlarna ska utvecklas. För att dra slutsatser om de atmosfäriska virvelgatorna har 11 satellitbilder med virvelgator analyserats. Sambandet λ = 3.9b-5.3 (förklaringsgrad r2 = 0.91) hittades mellan virvelgatans våglängd λ och bredden b på ön. Kvoten λ/b beräknades till medelvärdet 4.33 vilket är jämförbart med resultat från en liknande studie. / The visual appearance of the atmospheric vortex street behind a high mountain on an island is very similar to the periodic pattern caused by the flow past a circular cylinder. The friction between the fluid and the surface of the cylinder creates a wake downstream of the cylinder and periodic von Kármán vortex shedding occurs. The flow may be characterized by means of the Reynolds number and using a numerical model the different states can be simulated. If the algorithm lacks a turbulence model, the results for high Reynolds numbers will be wrong. The atmospheric vortex streets do not, however, arise due to friction. Instead, blocking of air masses and density variations are needed for the vortices to develop. To be able to draw conclusions about atmospheric vortex streets 11 satellite images showing the vortex streets have been analyzed. The relation λ = 3.9b-5.3 (coefficient of determination r2 = 0.91) was found, where λ is the wavelength of the vortex street and b the width of the island. The mean value of the ratio λ/b is 4.33 which is comparable with results from a similar study.

With the advent of very high spatial resolution (VHR) satellite, spatial details within the image scene have increased considerably. This led to the development of object-based image analysis (OBIA) for the analysis of VHR satellite images. Image segmentation is the fundamental step for OBIA. However, a large number of techniques exist for RS image segmentation. To identify the best ones for VHR imagery, a comprehensive literature review on image segmentation is performed. Based on that review, it is found that the multiresolution segmentation, as implemented in the commercial software eCognition, is the most widely-used technique and has been successfully applied for wide variety of VHR images. However, the multiresolution segmentation suffers from the parameter estimation problem. Therefore, this study proposes a solution to the problem of the parameter estimation for improving its efficiency in VHR image segmentation. The solution aims to identify the optimal parameters, which correspond to optimal segmentation. The solution to the parameter estimation is drawn from the Equations related to the merging of any two adjacent objects in multiresolution segmentation. The solution utilizes spectral, shape, size, and neighbourhood relationships for a supervised solution. In order to justify the results of the solution, a global segmentation accuracy evaluation technique is also proposed. The solution performs excellently with the VHR images of different sensors, scenes, and land cover classes. In order to justify the applicability of solution to a real life problem, a building detection application based on multiresolution segmentation from the estimated parameters, is carried out. The accuracy of the building detection is found nearly to be eighty percent. Finally, it can be concluded that the proposed solution is fast, easy to implement and effective for the intended applications.

The discontinuities play an important role both in design and development stages of many geotechnical engineering projects. Because of that considerable time and capital should be spent to determine discontinuity sets by conventional methods. This thesis present the results of the studies associated with the application of the Remote Sensing (RS) and the development of a methodology in accurately and automatically detecting the discontinuity sets. For detection of the discontinuities, automatic lineament analysis is performed by using high resolution satellite imagery for identification of rock discontinuities. The study area is selected as an Andesite quarry area in G&ouml / lbaSi, Ankara, Turkey. For the high resolution data 8-bit Ikonos Precision Plus with 1 meter resolution orthorectified image is used. The automatic lineament extraction process is carried out with LINE module of PCI Geomatica v8.2. In order to determine the most accurate parameters of LINE, an accuracy assessment is carried out. To be the reference of the output, manual lineament extraction with directional filtering in four principal directions (N-S, E-W, NE-SW, NW-SE) is found to be the most suitable method. For the comparison of automatic lineament extraction and manual lineament extraction processes, LINECOMP program is coded in java environment. With the written code, a location and length based accuracy assessment is carried out. After the accuracy assesssment, final parameters of automatically extracted lineaments for rock discontinuity mapping for the study area are determined. Besides these, field studies carried out in the study area are also taken into consideration.

The human vision system is generally recognized as being superior to all known artificial vision systems. Visual attention, among many processes that are related to human vision, is responsible for identifying relevant regions in a scene for further processing. In most cases, analyzing an entire scene is unnecessary and inevitably time consuming. Hence considering visual attention might be advantageous. A subfield of computer vision where this particular functionality is computationally emulated has been shown to retain high potential in solving real world vision problems effectively. In this monograph, elements of visual attention are explored and algorithms are proposed that exploit such elements in order to enhance image understanding capabilities. Satellite images are given special attention due to their practical relevance, inherent complexity in terms of image contents, and their resolution. Processing such large-size images using visual attention can be very helpful since one can first identify relevant regions and deploy further detailed analysis in those regions only. Bottom-up features, which are directly derived from the scene contents, are at the core of visual attention and help identify salient image regions. In the literature, the use of intensity, orientation and color as dominant features to compute bottom-up attention is ubiquitous. The effects of incorporating an entropy feature on top of the above mentioned ones are also studied. This investigation demonstrates that such integration makes visual attention more sensitive to fine details and hence retains the potential to be exploited in a suitable context. One interesting application of bottom-up attention, which is also examined in this work, is that of image segmentation. Since low salient regions generally correspond to homogenously textured regions in the input image; a model can therefore be learned from a homogenous region and used to group similar textures existing in other image regions. Experimentation demonstrates that the proposed method produces realistic segmentation on satellite images. Top-down attention, on the other hand, is influenced by the observer’s current states such as knowledge, goal, and expectation. It can be exploited to locate target objects depending on various features, and increases search or recognition efficiency by concentrating on the relevant image regions only. This technique is very helpful in processing large images such as satellite images. A novel algorithm for computing top-down attention is proposed which is able to learn and quantify important bottom-up features from a set of training images and enhances such features in a test image in order to localize objects having similar features. An object recognition technique is then deployed that extracts potential target objects from the computed top-down attention map and attempts to recognize them. An object descriptor is formed based on physical appearance and uses both texture and shape information. This combination is shown to be especially useful in the object recognition phase. The proposed texture descriptor is based on Legendre moments computed on local binary patterns, while shape is described using Hu moment invariants. Several tools and techniques such as different types of moments of functions, and combinations of different measures have been applied for the purpose of experimentations. The developed algorithms are generalized, efficient and effective, and have the potential to be deployed for real world problems. A dedicated software testing platform has been designed to facilitate the manipulation of satellite images and support a modular and flexible implementation of computational methods, including various components of visual attention models.

<p>Climate change is recognized as one of the most significant challenge of the 21st century. Anthropogenic activities have led to a substantial increase in greenhouse gases (GHGs) since the Industrial Revolution, with the energy sector being one the biggest contributors globally. The energy sector is now facing unique challenges not only due to decarbonization goals but also due to increased risks of climate extremes under climate change. </p><p>This dissertation focuses on leveraging machine learning, specifically utilizing unstructured data such as images, to address many of the unprecedented challenges faced by the energy systems. The dissertation begins (Chapter 1) by providing an overview of the risks posed by climate change to modern energy systems. It then explains how machine learning applications can help with addressing these risks. By harnessing the power of machine learning and unstructured data, this research aims to contribute to the development of more resilient and sustainable energy systems, as described briefly below. </p><p>Accurate forecasting of generation is essential for mitigating the risks associated with the increased penetration of intermittent and non-dispatchable variable renewable energy (VRE). In Chapters 2 and 3, deep learning techniques are proposed to predict solar irradiance, a crucial factor in solar energy generation, in order to address the uncertainty inherent in solar energy. Specifically, Chapter 2 introduces a cost-efficient fully exogenous solar irradiance forecasting model that effectively incorporates atmospheric cloud dynamics using satellite imagery. Building upon the work of Chapter 2, Chapter 3 extends the model to a fully probabilistic framework that not only forecasts the future point value of irradiance but also quantifies the uncertainty of the prediction. This is particularly important in the context of energy systems, as it relates to high-risk decision making.</p><p>While the energy system is a major contributor to GHG emissions, it is also vulnerable to climate change risks. Given the essential role of energy systems infrastructure in modern society, ensuring reliable and sustainable operations is of utmost importance. However, our understanding of reliability analysis in electricity transmission networks is limited due to the lack of access to large-scale transmission network topology datasets. Previous research has mostly relied on proxy or synthetic datasets. Chapter 4 addresses this research gap by proposing a novel deep learning-based object detection method that utilizes satellite images to construct a comprehensive large-scale transmission network dataset.</p>

As a strategy to conserve tropical rainforests of the East Usambara block of the Eastern Arc Mountains, Tanzania, I developed a set of models that can identify above-average tree species richness areas within the humid forests. I developed the model based on geo-referenced field data and satellite image-based variables from the Amani Nature Reserve, the largest forest sector in the East Usambara. I then verified the model by applying it to the Nilo Forest Reserve. The field data, part of the Tanzanian National Biodiversity Database, were collected by Frontier-Tanzania between 1999 and 2001, through the East Usambara Conservation Area Management Program, Government of Tanzania. The field data used are rapidly collectible by people with varied backgrounds and education. I gathered spectral reflectance values from pixels in the Landsat Enhanced Thematic Mapper (Landsat ETM) image covering the study area that corresponded to the ground sample points. The spectral information from different bands formed the satellite image-based variables in the dataset. The best satellite image logistic regression and discriminant analysis models were based on a single band, raw Landsat ETM mid-infrared band 7 (RB7). In the Amani forest, the RB7-based model resulted in 65.3% overall accuracy in identifying above average tree species locations. When the logistic and discriminant models were applied to Nilo forest sector, the overall accuracy was 62.3%. Of the rapidly collectible field variables, only tree density (number of trees) was selected in the logistic regression and the discriminant analysis models. Logistic and discriminant models using both RB7 and number of trees recorded 76.3% overall accuracy in Amani, and when applied to Nilo, 76.8% accuracy. It is possible to apply and adapt the current set of models to identify above-average tree species richness areas in East Usambara and other forest blocks of the Eastern Arc Mountains. Potentially, managers and researchers can periodically use the model to rapidly assess, monitor, update, and map the tree species rich areas within the forest. The same or similar models could be applied to check their applicability in other humid tropical forest areas. / Ph. D.

Satellite remote sensing has revolutionized the acquisition of large amounts of data, employing both active and passive sensors to capture critical information about our planet. These data can be analysed by using deep learning methodologies that demonstrate excellent capabilities in extracting the semantics from the data. However, one of the main challenges in exploiting the power of deep learning for remote sensing applications is the lack of labeled training data. Deep learning architectures, typically demand substantial quantities of training samples to achieve optimal performance. Motivated by the above-mentioned challenges, this thesis focuses on the limited availability of labeled datasets. These challenges include issues such as ambiguous labels in case of large-scale remote sensing datasets, particularly when dealing with the analysis of multi-source satellite remote sensing images. By employing novel deep learning techniques and cutting-edge methodologies, this thesis endeavors to contribute to advancements in the field of remote sensing. In this thesis, the problems related to limited labels are solved in several ways by developing (i) a novel spectral index generative adversarial network to augment real training samples for generating class-specific remote sensing data to provide a large number of labeled samples to train a neural-network classifier; (ii) a mono- and dual-regulated contractive-expansive-contractive convolutional neural network architecture to incorporate spatial-spectral information of multispectral data and minimize the loss in the feature maps and extends this approach to the analysis of hyperspectral images; (iii) a hybrid deep learning architecture with a discrete wavelet transform and attention mechanism to deal with few labeled samples for scene-based classification of multispectral images; and (iv) a weak supervised semantic learning technique that utilises weak or low-resolution labeled samples with multisource remote sensing images for predicting pixel-wise land-use-land-cover maps. The experiments show that the proposed approaches perform better than the state-of-the-art methods on different benchmark datasets and in different conditions.

Les travaux présentés dans cette thèse concernent l’utilisation de données de télédétection inédites pour le suivi des glaciers du massif du Mont Blanc : les images radar à synthèse d’ouverture Haute Résolution (HR) du satellite TerraSAR-X et les prises de vue HR d’un appareil photo numérique automatique. Cette thèse s’attache à montrer l’apport de ces sources d’informations hétérogènes pour mesurer le déplacement de surface des glaciers alpins. Dans cette optique, un examen des méthodes de mesure de déplacement spécifiques à chacun des deux types d’images est réalisé. Deux approches sont alors explorées : la mesure de déplacement monosource dans la géométrie propre à chaque capteur et la mesure de déplacement multisource via des combinaisons intra-capteur et inter-capteur. Alors que l’approche monosource fournit uniquement des mesures 2D du déplacement, les mesures multisources permettent pour la première fois d’estimer des champs de déplacement 3D de la surface des glaciers du Mont Blanc. Les mesures ont été réalisées sur plusieurs séries temporelles d’images couvrant la période 2008-2009 pour quatre glaciers du massif du Mont Blanc (Argentière, Mer de Glace/Leschaux, Bossons et Taconnaz). Dans le contexte du changement climatique, ces mesures de déplacement de surface fournissent une donnée intéressante en glaciologie pour contraindre les modèles numériques d’écoulement et d’évolution des glaciers. / The works presented in this PhD thesis focuses on the use of new remote sensing data for "massif du Mont Blanc" glaciers’ monitoring: High Resolution (HR) synthetic aperture radar images of TerraSAR-X satellite and HR shooting of the automatic digital camera. This thesis will show the contribution of this heterogeneous information to the measurement of the surface displacement of alpine glacier. For this purpose, a review of displacement measurement methods specific to each of the two types of image is proposed. Then two approaches are explored: the mono-source displacement measurement in the geometry for each sensor and the multi-source displacement measurement via intra-sensor and inter-sensor combinations. While the mono-source approach provides only 2D displacement measurements, multi-source measurements allow, for the first time, the estimation of 3D surface displacement fields of the Mont Blanc glaciers. The measurements were performed on several image time series covering the period 2008-2009 for four Mont Blanc glaciers (Argentière, Mer de Glace/Leschaux, Bossons et Taconnaz). In the context of global warming, these surface displacement measurements provide interesting data in the glaciology domain in order to constrain flow and evolution digital models.