No description available.

Any generic computer vision algorithm must be able to copewith the variations in appearance of objects due to differentillumination conditions. While these variations in the shadingof a surface may seem a nuisance, they in fact containinformation about the world. This thesis tries to provide anunderstanding what information can be extracted from theshading in a single image and how to achieve this. One of thechallenges lies in finding accurate models for the wide varietyof conditions that can occur. Frequency space representations are powerful tools foranalyzing shading theoretically. Surfaces act as low-passfilters on the illumination making the reflected lightband-limited. Hence, it can be represented by a finite numberof components in the Fourier domain, despite having arbitraryillumination. This thesis derives a basis for shading byrepresenting the illumination in spherical harmonics and theBRDF in a basis for isotropic reflectance. By analyzing thecontributing variance of this basis it is shown how to createfinite dimensional representations for any surface withisotropic reflectance. The finite representation is used to analytically derive aprincipal component analysis (PCA) basis of the set of imagesdue to the variations in the illumination and BRDF. The PCA isperformed model-based so that the variations in the images aredescribed by the variations in the illumination and the BRDF.This has a number of advantages. The PCA can be performed overa wide variety of conditions, more than would be practicallypossible if the images were captured or rendered. Also, thereis an explicit mapping between the principal components and theillumination and BRDF so that the PCA basis can be used as aphysical model. By combining a database of captured illumination and adatabase of captured BRDFs a general basis for shading iscreated. This basis is used to investigate materialclassification from a single image with known geometry butarbitrary unknown illumination. An image is classified byestimating the coecients in this basis and comparing them to adatabase. Experiments on synthetic data show that materialclassification from reflectance properties is hard. There aremis-classifications and the materials seem to cluster intogroups. The materials are grouped using a greedy algorithm.Experiments on real images show promising results. Keywords:computer vision, shading, illumination,reflectance, image irradiance, frequency space representations,spherical harmonics, analytic PCA, model-based PCA, materialclassification, illumination estimation

<p>Any generic computer vision algorithm must be able to copewith the variations in appearance of objects due to differentillumination conditions. While these variations in the shadingof a surface may seem a nuisance, they in fact containinformation about the world. This thesis tries to provide anunderstanding what information can be extracted from theshading in a single image and how to achieve this. One of thechallenges lies in finding accurate models for the wide varietyof conditions that can occur.</p><p>Frequency space representations are powerful tools foranalyzing shading theoretically. Surfaces act as low-passfilters on the illumination making the reflected lightband-limited. Hence, it can be represented by a finite numberof components in the Fourier domain, despite having arbitraryillumination. This thesis derives a basis for shading byrepresenting the illumination in spherical harmonics and theBRDF in a basis for isotropic reflectance. By analyzing thecontributing variance of this basis it is shown how to createfinite dimensional representations for any surface withisotropic reflectance.</p><p>The finite representation is used to analytically derive aprincipal component analysis (PCA) basis of the set of imagesdue to the variations in the illumination and BRDF. The PCA isperformed model-based so that the variations in the images aredescribed by the variations in the illumination and the BRDF.This has a number of advantages. The PCA can be performed overa wide variety of conditions, more than would be practicallypossible if the images were captured or rendered. Also, thereis an explicit mapping between the principal components and theillumination and BRDF so that the PCA basis can be used as aphysical model.</p><p>By combining a database of captured illumination and adatabase of captured BRDFs a general basis for shading iscreated. This basis is used to investigate materialclassification from a single image with known geometry butarbitrary unknown illumination. An image is classified byestimating the coecients in this basis and comparing them to adatabase. Experiments on synthetic data show that materialclassification from reflectance properties is hard. There aremis-classifications and the materials seem to cluster intogroups. The materials are grouped using a greedy algorithm.Experiments on real images show promising results.</p><p><b>Keywords:</b>computer vision, shading, illumination,reflectance, image irradiance, frequency space representations,spherical harmonics, analytic PCA, model-based PCA, materialclassification, illumination estimation</p>

Comprendre les variations de l’irradiance solaire, notamment dans le domaine UV, est essentiel pour les modèles climatiques. Les modèles d’irradiance sont précieux pour reconstruire l'irradiance solaire spectrale (SSI) en l'absence d'observations ou lorsque celles-ci manquent de stabilité. Cependant, ils font certaines hypothèses sur les structures solaires. Nous visons ici à contraindre ces hypothèses en caractérisant le contraste UV des structures solaires. Grâce aux données du satellite Solar Dynamic Observatory (SDO)entre 2010 et 2016, nous quantifions pour la première fois les contrastes dans l'UV. L'étude du contraste des structures solaires et de leur segmentation montre que des seuils photométriques sont nécessaires pour segmenter correctement les structures solaires, principalement en UV, en raison de la coexistence de structures sombres et brillantes pour la même valeur du champ magnétique. Certains pixels classés parmi le Soleil calme par le modèle SATIRE-S appartiennent en fait aux facules, mais ils sont trop peu nombreux pour avoir un impact sur les reconstructions de SSI. Nos résultats soulignent l'importance des observations multi-longueurs d'onde pour mieux contraindre l'identification des structures. Distinguer réseau et facule est essentiel pour reconstruire la SSI sur une longue période, et la prise en compte de la dépendance du contraste du réseau par rapport au champ magnétique améliore la reconstruction de la SSI. Enfin, nous ne trouvons aucun indice de variations du contraste durant le cycle solaire.Nous présentons aussi des résultats sur les corrélations entre les émissions Hα et Ca II des étoiles de type solaire. Nous montrons que l’hypothèse de Meunier et al., 2009 pour expliquer les anti-corrélations de certaines étoiles est confirmée par nos résultats. / Understanding solar irradiance variations,in particular in the ultraviolet wavelength range, is essential for climate modelling. Solar irradiance models are precious for reconstructing the spectral solar irradiance (SSI) in the absence of observations or when they lack stability. However, they come with their assumptions. Here we aim here to constrain these in the UV by characterising the contrast of solar magnetic features in the UV.From solar images taken by the Solar Dynamic Observatory (SDO) between 2010 and 2016, we quantify UV contrasts the first time. The study of the contrast of the solar structures and their segmentation shows that photometric thresholds are necessary to properly segment solar structures, mainly in the UV, because of the coexistence of both dark and bright structures for the same value of the magnetic field. Some pixels that are classified as quiet-Sun by the SATIRE-S model actually belong to faculae, but they are too few to have a significant impact on SSI reconstructions. Our results highlight the importance of multi-wavelength observations for better constraining the identification of structures. Distinguishing network and faculae is essential for such reconstructions over a long period, and using a network with magnetically variable contrast improves SSI reconstruction, Finally, we find no evidence of contrast variations during the solar cycle.We also present results on the correlations between Hα and Ca II emissions of Sun-like stars. We show that the hypothesis of Meunier et al., 2009 to explain the anti-correlations of certain stars corresponds to our observations.

L’irradiance spectrale solaire (SSI) dans la bande ultraviolette est un paramètre-clé pour la spécification de la moyenne et la haute atmosphère terrestre. Elle est requise dans de nombreuses applications en météorologie de l’espace, et aussi pour l’étude du climat. Or les observations souffrent de plusieurs défauts : manque de couverture spectrale et temporelle permanente, dégradation des capteurs, désaccords entre les instruments, etc. Plusieurs modèles de reconstruction de la SSI ont été développés pour pallier à ces difficultés. Chacun souffre de défauts, et la reconstruction du spectre en-dessous de 120nm est un réel défi. C’est dans ce contexte que nous avons développé un modèle empirique, qui recourt au champ magnétique photosphérique pour reconstruire les variations du spectre solaire. Ce modèle décompose les magnétogrammes solaires en différentes structures qui sont classées à partir de leur aire (et non sur la base de leur intensité, comme dans la plupart des autres modèles). La signature spectrale de ces structures est déduite des observations, et non pas imposée par des modèles de l’atmosphère solaire. La qualité de la reconstruction s’avère être comparable à celle d’autres modèles. Parmi les principaux résultats, relevons que deux classes seulement de structures solaires suffisent à reproduire correctement la variabilité spectrale solaire. En outre, seule une faible résolution radiale suffit pour reproduire les variations de centre-bord. Enfin, nous montrons que l’amélioration apportée par la décomposition du modèle en deux constantes de temps peut être attribuée à l’effet des raies optiquement minces. / The spectrally-resolved radiative output of the Sun (SSI) in the UV band, i.e. at wavelengths below 300 nm, is a key quantity for specifying the state of the middle and upper terrestrial atmosphere. This quantity is required in numerous space weather applications, and also for climate studies. Unfortunately, SSI observations suffer from several problems : they have numerous spectral and temporal gaps, instruments are prone to degradation and often disagree, etc. This has stimulated the development of various types of SSI models. Proxy-based models suffer from lack of the physical interpretation and are as good as the proxies are. Semi-empirical models do not perform well below 300 nm, where the local thermodynamic equilibrium approximation does not hold anymore. We have developed an empirical model, which assumes that variations in the SSI are driven by solar surface magnetic flux. This model proceeds by segmenting solar magnetograms into different structures. In contrast to existing models, these features are classified by their area (and not their intensity), and their spectral signatures are derived from the observations (and not from models). The quality of the reconstruction is comparable to that of other models. More importantly, we find that two classes only of solar features are required to properly reproduce the spectral variability. Furthermore, we find that a coarse radial resolution suffices to account for geometrical line-of-sight effects. Finally, we show how the performance of the model on different time-scales is related to the optical thickness of the emission lines.

Firstly, this dissertation proposes a new characterization and classification method for daily sky conditions by using the daily sky clearness index (KD) and the daily probability of persistence (POP-KD) that can be derived from ground-based irradiance measurement data. Quality of daily solar irradiance is characterized by a newly proposed parameter, POP-KD. This characterized daily quality is varying and uncertain at the middle level of the quantity, but high and more certain at very high and low quantity levels. In addition, the proposed characterization method shows interesting results for KD and POP-KD: a statistical consistency for multiple years and similarity for their seasonal trends. The classification results also indicate an existence of dominant classes, and transitions between the dominant classes are significant for all locations. This dissertation also generates annual synthetic sequences of KD and POP-KD using a Markov approach. The generated sequences show statistical similarities with observed sequences. Secondly, this dissertation proposes methodologies to estimate day-ahead solar irradiance using the National Weather Service (NWS) sky cover forecast. For model development, this paper splits up a direct estimation process from the sky cover forecast to solar irradiance into two stages: forecast verification and cloud-to-irradiance conversion. Uncertainty for each stage and for the overall estimation process is quantified. NWS forecast uncertainty (about 20%) is identified as the main source of uncertainty for the overall process. In addition, verification of the sky cover forecast shows approximately 20% overestimated bias at days with a high irradiance level. Thus, the NWS sky cover forecast needs to be adjusted based on the type of day. This dissertation also proposes a conversion equation relating daily quantity of cloud information and daily quantity of solar irradiance. The proposed conversion equation achieves accuracy with simplicity. Five day-ahead solar irradiance quantity estimation methods are proposed in this dissertation. The proposed methods incorporate different schemes for dealing with the bias discovered in the cloud forecast. The observed data are regularly found within the 95% confidence intervals of the estimated values. Estimation results demonstrate the effectiveness of the conditional adjustment schemes at different irradiance levels. Lastly, this dissertation proposes a methodology to estimate day-ahead solar irradiance using fluctuation information of the NWS sky cover forecast. POP-KD was used as a parameter for the quality of daily solar irradiance. POP-KD efficiently represents the quality of daily solar irradiance. In addition, POP-KD indicates the probability that solar irradiance variability is within the ramp rates of common generators in power systems at a certain photovoltaic penetration level. This dissertation also proposes a new equation for the conversion from cloud fluctuation information to daily quality of surface solar irradiance. The proposed equation achieves accuracy. The proposed day-ahead solar irradiance quality estimation method is based on fluctuation information provided by the NWS sky cover forecast. This method uses a normalization approach to relate fluctuation of cloud forecast and fluctuation of cloud observation. The observed data are regularly found within the 95% CIs of the estimated values. / Ph. D.

The European Water Framework Directive (WFD) has encouraged considerable research on the development of water quality bioindicators. Seagrasses, that are highly sensitive to direct and indirect anthropogenic stress, and specified as quality elements from the WFD, have been at the center of this effort. In this study the use of Cymodocea nodosa, a widely distributed angiosperm in the Mediterranean Sea, as a bioindicator of anthropogenic stress was tested. Key biotic features of two meadows growing in locations of contrasting ecological status in the N. Aegean Sea, Greece, were sampled and analysed following a hierarchical designed approach. Plants from the degraded meadow (Nea Karvali) were found to have significantly (p<0.05) longer leaves, higher N and P (%) content and lower C/N ratio in their leaves than the less degraded-pristine (Brasidas and/or Thasos) meadows. The application of chlorophyll fluorescence as an easily measurable indicator of the anthropogenic stress has been tested before with limited success. This study, based on a large amount of measurements (ca.500 per meadow) carried out after a short acclimation period in the laboratory under constant temperature and irradiance conditions, showed significantly higher (p<0.05) ΔF/Fm’ and Fm values at plants from the degraded than from the pristine meadows. Three sets of laboratory 8-days experiments were carried out under optimal growth temperature (21±1.5oC) in order to investigate cause-effects relationships between the main local stressors (nutrients-N, P, irradiance and heavy metal-Cu) to shoots collected from differently impacted meadows. High nutrient concentrations (30μΜ Ν-ΝΟ3; 2μM P-PO4) had a significant effect (p<0.05) on ΔF/Fm’ only on shoots from the pristine site. Low irradiance (37 μmol photons m-2 s-1) resulted in a significant increase (p<0.05) of ΔF/Fm’, while high copper concentrations (>4.7 μM) had the opposite effect. Through these experiments light availability and nutrients were identified as the main factor that affects the meadows health.

Several factors impact the light irradiance and relative illumination produced by a lens system at its image plane. In addition to the cosine-fourth-power radiometric law, image and pupil aberrations, and light vignetting also count. In this paper, we use an irradiance transport equation to derive a closed form solution that provides insight into how individual aberration terms affect the light irradiance and relative illumination. The theoretical results are in agreement with real ray tracing.

Several factors impact the light irradiance and relative illumination produced by a lens system at its image plane. In addition to cosine-fourth-power radiometric law, image and pupil aberrations and light vignetting also count. We use an irradiance transport equation to derive a closed form solution that provides insight into how individual aberration terms affect the light irradiance and relative illumination. The theoretical results are in agreement with real ray tracing. (C) 2016 Society of Photo-Optical Instrumentation Engineers (SPIE)

In recent years, different renewable energy sources have been on the rise. Among these, solar power has shown great potential in both small and big scale. Since solar irradiance is the main input for solar power systems, it is of great importance to examine how much of the solar irradiance actually reaches a certain location, and how much of the total solar irradiance consists of direct (beam), diffuse and reflected irradiance. This is usually done with expensive measurement meters, such as pyranometers and pyrheliometers. In this thesis, incident solar irradiance data from a cheap sensor network is analyzed, a new proposed model for estimating beam and diffuse fraction is examined and the results are compared to another estimation model, namely Erbs model. The comparison between the two models shows high correlation for beam irradiance, regardless of weather conditions, whereas the correlation for diffuse irradiance shows a highly varying result, much dependent on the weather conditions. It is difficult to motivate how well the proposed model is performing on a broader scale, since the study is limited to a specific area during a short period of time.