Dissertations / Theses on the topic 'Multi-scale classifications'

Cette thèse s'inscrit dans le domaine biomédical, à l'interface entre l'instrumentation et le traitement du signal. L'objectif de ce travail est d'obtenir, grâce à une mélange de différentes informations, un monitorage de l'activité du fœtus (rythme cardiaque et mouvements fœtaux) pour apprécier son état de bien-être ou de souffrance, et ceci aux différents stades de la grossesse. Actuellement, les paramètres qui caractérisent la souffrance fœtale, issus du rythme cardiaque et des mouvements fœtaux, sont évalués par le médecin et ils sont réunis dans le score de Manning. Deux inconvénients majeurs existent: a) l'évaluation du score est trop longue puisqu'elle dure 1 heure; b) il existe des variations inter et intra-opérateur conduisant à différentes interprétations du bilan médical de la patiente. Pour s'affranchir de ces désavantages nous évaluons le bien-être fœtal d'une façon objective, à travers le calcul d'un score. Pour atteindre ce but, nous avons développé une technologie ultrasonore multi-capteurs (12 capteurs) permettant de recueillir une soixantaine de (paires de) signaux Doppler en provenance du cœur, des membres inférieurs et supérieurs. Dans cette thèse notre première contribution s'appuie sur la mise en œuvre de nouveaux algorithmes de détection du rythme cardiaque (mono-canal et multi-canaux). Notre deuxième contribution concerne l'implémentation de deux catégories de paramètres basés sur le rythme cardiaque : a) la classes des paramètres "traditionnels" utilisés par les obstétriciens et évalués aussi dans le test de Manning (ligne de base, accélérations, décélérations); b) la classe des paramètres utilisés par les cardiologues et qui caractérisent la complexité d'une série temporelle (entropie approximée, entropie échantillonnée, entropie multi-échelle, graphe de récurrence, etc.). Notre troisième contribution consiste également à apporter des modifications notables aux différents algorithmes du calcul des mouvements fœtaux et des paramètres qui en découlent comme : le nombre de mouvements, le pourcentage du temps passé en mouvement par un fœtus, la durée des mouvements. Notre quatrième contribution concerne l'analyse conjointe du rythme cardiaque et des mouvements fœtaux. Cette analyse nous conduit alors à l'identification de différents états comportementaux. Le développement ou le non-développement de ces états est un indicateur de l'évolution neurologique du fœtus. Nous proposons d'évaluer les paramètres de mouvements pour chaque état comportemental. Enfin, notre dernière contribution concerne la mise en œuvre de différents scores et des classifications qui en découlent. Les perspectives directes à donner à ce travail concernent l'intégration des scores ou paramètres les plus pertinents dans un dispositif de surveillance à domicile ou bien dans un dispositif de monitorage clinique.

In the present thesis the crack propagation in the ductile-brittle transition region is studied on two scales with deterministic models. In the macroscopic model the ductile failure is described by a non-local Gurson-model whereas the discrete void microstructure is resolved around the crack tip in the microscopic model. The failure by cleavage is not evaluated by means of a post-processing criterion but is modeled equivalently using a cohesive zone model on both scales. Thus, cleavage is not a priori identified with unstable crack propagation but the transition between stable and unstable mode of propagation is a result of the simulation. The problem of handling completely failed material within the framework of non-local damage models is pointed out. A method to overcome this problem is proposed and successfully applied. The case of contained plastic yielding at the crack tip is addressed with a modified-boundary layer model. The macroscopic simulations reproduce many features which are known from experiments like the formation of stretch zones, cleavage after initial ductile tearing, pop-ins with crack arrest, among others. The microscopic simulations substantiate the understanding of the macroscopically observed behavior. Systematic parameter studies are performed. Starting with considerations on the limit cases like pure ductile failure or the lower-ductile brittle transition region allows to separate the effects of the different constitutive parameters. Based on these results, a methodology is proposed to extract the macroscopic material parameters from experiments. This scheme is successfully applied to experimental data from literature. The results show that the behavior of a low-constraint specimen can be reliably predicted with the parameters extracted from a high-constraint specimen<br>In der vorliegenden Arbeit wird die Rissausbreitung im spröd-duktilen Übergangsbereich auf zwei Skalen mittels deterministischer Modelle untersucht. Das duktile Versagen wird im makroskopischen Modell durch ein nichtlokales Gurson-Modell beschrieben, während im mikroskopischen Modell die Porenmikrostruktur im Bereich um die Rissspitze diskret aufgelöst wird. Das mögliche Versagen durch Spaltbruch wird nicht, wie üblich, nachträglich durch ein spannungsbasiertes Kriterium bewertet. Stattdessen wird der Spaltbruch auf beiden Skalen durch ein Kohäsivzonenmodell abgebildet. Somit wird die Spaltbruchinitiierung nicht a priori mit instabiler Rissausbreitung gleichgesetzt. Vielmehr ist die Stabilität der Rissausbreitung ein Ergebnis der Simulationen. Außerdem wird das Problem der der Handhabung vollständig ausgefallenen Materials im Rahmen nichtlokaler Schädigungsmodelle herausgestellt. Es wird eine Methode vorgestellt, dieses Problem zu behandeln und erfolgreich angewendet. In den Simulationen wird der Fall vollständig eingebetteten, plastischen Fließens untersucht. Die Simulationen mit dem makroskopischen Modell geben viele Effekte wieder, die aus Experimenten bekannt sind. Dazu zählen die Ausbildung von Stretchzonen, die Spaltbruchinitiierung nach anfänglichem, duktilem Reißen oder lokale Instabilitäten mit Rissarrest. Die mikroskopischen Simulationen tragen zum Verständnis des makroskopisch beobachteten Verhaltens bei. In der vorliegenden Arbeit werden systematische Parameterstudien durchgeführt. Zunächst werden Grenzfälle wie das rein duktile Versagens oder der Spaltbruch in Abwesenheit der Mikroporen untersucht, um die Einflüsse der einzelnen Materialparameter abzugrenzen. Ausgehend von diesen Ergebnissen wird eine Prozedur vorgeschlagen, die Materialparameter des makroskopischen Modells Schritt für Schritt aus Experimenten zu bestimmen. Diese Prozedur wird erfolgreich auf experimentelle Daten aus der Literatur angewendet. Die Ergebnisse zeigen, dass es das entwickelte Modell erlaubt, das Verhalten einer Bruchmechanikprobe mit geringer Dehnungsbehinderung an der Rissspitze mit denjenigen Materialparametern vorherzusagen, die an Proben mit einer hohen Dehnungsbehinderung ermittelt wurden

Cette thèse explore le potentiel de l'apprentissage automatique pour améliorer la cartographie de modèles complexes d'utilisation des sols et de la couverture terrestre à l'aide de données d'observation de la Terre. Traditionnellement, les méthodes de cartographie reposent sur la classification et l'interprétation manuelles des images satellites, qui sont sujettes à l'erreur humaine. Cependant, l'application de l'apprentissage automatique, en particulier par le biais des réseaux neuronaux, a automatisé et amélioré le processus de classification, ce qui a permis d'obtenir des résultats plus objectifs et plus précis. En outre, l'intégration de données de séries temporelles d'images satellitaires (STIS) ajoute une dimension temporelle aux informations spatiales, offrant une vue dynamique de la surface de la Terre au fil du temps. Ces informations temporelles sont essentielles pour une classification précise et une prise de décision éclairée dans diverses applications. Les informations d'utilisation des sols et de la couverture terrestre précises et actuelles dérivées des données STIS sont essentielles pour guider les initiatives de développement durable, la gestion des ressources et l'atténuation des risques environnementaux. Le processus de cartographie de d'utilisation des sols et de la couverture terrestre à l'aide du l'apprentissage automatique implique la collecte de données, le prétraitement, l'extraction de caractéristiques et la classification à l'aide de divers algorithmes l'apprentissage automatique . Deux stratégies principales de classification des données STIS ont été proposées : l'approche au niveau du pixel et l'approche basée sur l'objet. Bien que ces deux approches se soient révélées efficaces, elles posent également des problèmes, tels que l'incapacité à capturer les informations contextuelles dans les approches basées sur les pixels et la complexité de la segmentation dans les approches basées sur les objets.Pour relever ces défis, cette thèse vise à mettre en œuvre une métho basée sur des informations multi-échelles pour effectuer la classification de l'utilisation des terres et de la couverture terrestre, en couplant les informations spectrales et temporelles par le biais d'une méthodologie combinée pixel-objet et en appliquant une approche méthodologique pour représenter efficacement les données multi-variées SITS dans le but de réutiliser la grande quantité d'avancées de la recherche proposées dans le domaine de la vision par ordinateur<br>In this thesis, the potential of machine learning (ML) in enhancing the mapping of complex Land Use and Land Cover (LULC) patterns using Earth Observation data is explored. Traditionally, mapping methods relied on manual and time-consuming classification and interpretation of satellite images, which are susceptible to human error. However, the application of ML, particularly through neural networks, has automated and improved the classification process, resulting in more objective and accurate results. Additionally, the integration of Satellite Image Time Series(SITS) data adds a temporal dimension to spatial information, offering a dynamic view of the Earth's surface over time. This temporal information is crucial for accurate classification and informed decision-making in various applications. The precise and current LULC information derived from SITS data is essential for guiding sustainable development initiatives, resource management, and mitigating environmental risks. The LULC mapping process using ML involves data collection, preprocessing, feature extraction, and classification using various ML algorithms. Two main classification strategies for SITS data have been proposed: pixel-level and object-based approaches. While both approaches have shown effectiveness, they also pose challenges, such as the inability to capture contextual information in pixel-based approaches and the complexity of segmentation in object-based approaches.To address these challenges, this thesis aims to implement a method based on multi-scale information to perform LULC classification, coupling spectral and temporal information through a combined pixel-object methodology and applying a methodological approach to efficiently represent multivariate SITS data with the aim of reusing the large amount of research advances proposed in the field of computer vision

Der Beitrag gibt einen kompakten Überblick zur Leistungsfähigkeit und Anwendbarkeit der in den Teilprojekten D2-Numerische Simulation, E3-Sicherheitsbeurteilung und E4-Numerische Langzeitprognose des Sonderforschungsbereichs 528 entwickelten Algorithmen und Programmlösungen. Die gezeigten Methoden sind praxistauglich aufbereitet und stehen zur Anwendung für die Analyse des Kurz- und Langzeit-Tragverhaltens von Textilbetonstrukturen zur Verfügung<br>The paper provides a compact summary of the ability and applicability of the algorithms and software packages developed in the project parts D2-Numerical Simulation, E3-Reliability Assessment und E4-Numerical Long-term Prognosis of the Collaborative Research Centre 528. The presented methods are prepared for practical use and are available for the analysis of the short- and longterm load-bearing behaviour of textile reinforced concrete structures

Doctor of Philosophy<br>This thesis is based on research on financial time series analysis using pattern recognition methods. The first part of this research focuses on univariate time series analysis using different pattern recognition methods. First, probabilities of basic patterns are used to represent the features of a section of time series. This feature can remove noise from the time series by statistical probability. It is experimentally proven that this feature is successful for pattern repeated time series. Second, a multiscale Gaussian gravity as a pattern relationship measurement which can describe the direction of the pattern relationship is introduced to pattern clustering. By searching for the Gaussian-gravity-guided nearest neighbour of each pattern, this clustering method can easily determine the boundaries of the clusters. Third, a method that unsupervised pattern classification can be transformed into multiscale supervised pattern classification by multiscale supervisory time series or multiscale filtered time series is presented. The second part of this research focuses on multivariate time series analysis using pattern recognition. A systematic method is proposed to find the independent variables of a group of share prices by time series clustering, principal component analysis, independent component analysis, and object recognition. The number of dependent variables is reduced and the multivariate time series analysis is simplified by time series clustering and principal component analysis. Independent component analysis aims to find the ideal independent variables of the group of shares. Object recognition is expected to recognize those independent variables which are similar to the independent components. This method provides a new clue to understanding the stock market and to modelling a large time series database.

This thesis is based on research on financial time series analysis using pattern recognition methods. The first part of this research focuses on univariate time series analysis using different pattern recognition methods. First, probabilities of basic patterns are used to represent the features of a section of time series. This feature can remove noise from the time series by statistical probability. It is experimentally proven that this feature is successful for pattern repeated time series. Second, a multiscale Gaussian gravity as a pattern relationship measurement which can describe the direction of the pattern relationship is introduced to pattern clustering. By searching for the Gaussian-gravity-guided nearest neighbour of each pattern, this clustering method can easily determine the boundaries of the clusters. Third, a method that unsupervised pattern classification can be transformed into multiscale supervised pattern classification by multiscale supervisory time series or multiscale filtered time series is presented. The second part of this research focuses on multivariate time series analysis using pattern recognition. A systematic method is proposed to find the independent variables of a group of share prices by time series clustering, principal component analysis, independent component analysis, and object recognition. The number of dependent variables is reduced and the multivariate time series analysis is simplified by time series clustering and principal component analysis. Independent component analysis aims to find the ideal independent variables of the group of shares. Object recognition is expected to recognize those independent variables which are similar to the independent components. This method provides a new clue to understanding the stock market and to modelling a large time series database.

The role of mangroves as essential fish habitat has been a focus of extensive research. However, recent evidence has shown that this role should not be evaluated in isolation from surrounding habitats such as seagrass beds and hard-bottom communities. For example, submerged aquatic vegetation (SAV) communities provide potential sources of food and shelter for fish species that may reside in the mangroves, but may also undergo ontogenetic migrations and daily home-range movements into neighboring habitats. The connectivity between the mangrove fish community and the surrounding seascape may be influenced by the level of patchiness, fragmentation, and spatial heterogeneity of adjacent SAV habitats (i.e., SAV seascape structure). The spatial patterns and heterogeneity of SAV seascape structures are driven by internal and external regulatory mechanisms operating at different spatial and temporal scales. In addition, it is likely that many fish species inhabiting the mangrove zones have different home ranges, and foraging and migratory patterns; therefore, different mangrove fish species may respond to seascape heterogeneity at different scales. There are few studies that have assessed the influence and connectivity of benthic habitats adjacent to mangroves for estuarine fish populations at multiple scales. The present research used an exploratory seascape approach in Biscayne Bay (Florida, USA) to evaluate patterns in the patch composition and configuration of SAV communities, and to examine relationships between seascape structural metrics and the abundance, diversity, and distribution of fishes that utilize the adjacent mangrove shoreline as nursery and/or adult habitat. This seascape approach consisted of: a) the multi-scale characterization of the SAV distribution across the seascape with metrics developed in Landscape Ecology, Geographic Information Systems and Remote Sensing; b) multivariate analyses to identify groups with significantly distinct SAV seascape structures within the most heterogeneous scale, and identify possible mechanisms driving the observed SAV seascape structures; and c) an assessment of the mangrove fish community responses to SAV seascape structures. By applying a set of multivariate analyses (e.g., ANOSIM, MDS plots, hierarchical clustering), the buffer within 200 m from shore was identified as the scale with the highest structural heterogeneity. At this scale, two major SAV seascape structures (i.e., areas with similar SAV spatial arrangement and composition) were identified: a fragmented SAV seascape (FSS) structure and a continuous SAV seascape (CSS) structure. Areas with CSS were characterized by large, uniform SAV patches. In contrast, areas with FSS were characterized by a higher density of smaller, more complex SAV patches. Furthermore, the areas with CSS and FSS structures clustered in zones of the bay with distinct salinity properties. The areas with CSS structures were mostly located in zones characterized by high and stable salinity. However, the areas with FSS concentrated in zones that are influenced by freshwater discharges from canals and with low and variable salinity. The responses of fish diversity metrics were not constrained to the scale at which the greatest spatial heterogeneity of SAV seascape structures was observed (i.e., the seascape composition and configuration within 200 m from shore), but was related to SAV seascape characteristics across different scales. The majority of the variability of the fish diversity metrics in the mangrove shoreline was explained by SAV seascape structures within the smaller scales (i.e., 100-400 m from shore), and SAV seascape structures that represented the level of fragmentation and/or the percent of suitable habitat. Different conceptual models were proposed to illustrate and understand the ecological dynamics behind the relationship between the diversity of the mangrove fish community and the structure of the adjacent SAV seascape. In general, the diversity and abundance of fishes is influenced by the type and level of fragmentation of the SAV seascape, which, in turn, influence the proportion of the seascape used for foraging and refuge by fish. In conclusion, this research quantified how the release of large pulses of freshwater into near-shore habitats of coastal lagoons can influence the seascape structure of SAV communities. Namely, freshwater inputs produce fragmentation in otherwise fairly homogeneous SAV meadows. The outcome of this research highlights the importance of seascape characteristics as indicators of ecosystem-level modifications and alterations affecting the spatial distribution, assemblage, and diversity of marine nearshore habitats in coastal regions heavily influenced by human activities. In addition, the results illustrated the cascading effects and synergistic influences of near-shore habitat spatial assemblages on the composition and diversity of estuarine fish communities. Lastly, and very importantly, the relationships established in this project provide quantitative and qualitative information on patterns of species-habitat associations needed for the improved synergistic management and protection of coastal habitats and fisheries resources.

Submitted by MARIA JACQUELINE ATOCHE BRAVO (jacqui_mab@hotmail.com) on 2016-05-13T13:28:28Z No. of bitstreams: 1 disertacao__Jacqui.pdf: 8482416 bytes, checksum: 2325158a94282088f873ac31bbd97305 (MD5)<br>Approved for entry into archive by Felipe Augusto Arakaki (arakaki@reitoria.unesp.br) on 2016-05-16T12:30:13Z (GMT) No. of bitstreams: 1 bravo_mja_me_sjrp.pdf: 8482416 bytes, checksum: 2325158a94282088f873ac31bbd97305 (MD5)<br>Made available in DSpace on 2016-05-16T12:30:13Z (GMT). No. of bitstreams: 1 bravo_mja_me_sjrp.pdf: 8482416 bytes, checksum: 2325158a94282088f873ac31bbd97305 (MD5) Previous issue date: 2016-03-31<br>Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)<br>No presente trabalho, apresenta-se uma revisão sobre os principais abordagens para análise e classificação de texturas, entre eles o descritor LBP (Local Binary Pattern), o descritor LFP (Local Fuzzy Patterm) e o descritor MSLMP (Multi-scale Local Mapped Pattern), o qual é uma extensão multiescalar do descritor LMP (Local Mapped Pattern). Resultados anteriores presentes na literatura, indicaram que o MSLMP conseguiu resultados superiores aos mencionados anteriormente. Neste trabalho propõe-se uma análise mais abrangente sobre sua viabilidade para concluir que o MSLMP é mais eficaz que os anteriores. Essa análise é feita alterando-se a Matriz de Pesos para os pixels limiarizados. Para avaliar seu desempenho, foi utilizada a base de texturas do Album de Brodatz. Após processá-la pelo descritor MSLMP, com cada uma das matrizes de Pesos propostas neste trabalho, foram comparadas as taxas de acertos alcançadas usando a distância Chi-quadrado. Resultados experimentais mostram um valor de sensibilidade melhor para o descritor MSLMP em comparação aos outros descritores presentes na literatura.<br>This work, presents a review about the main techniques for analysis and classification of textures, including the LBP descriptor (Local Binary Pattern), the descriptor LFP (Local Fuzzy Pattern) and the descriptor MSLMP (Multi-Scale Local Mapped Pattern), which is a multi-scale extension of the LMP method (Local Mapped Pattern). Previous results present in the literature, indicated that the MSLMP achieved better results than those mentioned above. This work proposes a more comprehensive analysis of its feasibility to conclude that this descriptor is more effective than the others. This analysis is done by changing the weight matrix for the thresholding pixels. To evaluate its performance, it was used the texture base of the Brodatz album. After processing it by the descriptor MSLMP with each of the weights matrices proposed in this work, the achieved hit rates were compared by using the distance Chi-square. Experimental results show a better sensitivity value for MSLMP descriptor in comparison of other descriptors present in the literature.<br>CNPq: 131632/2014-0

Die Arbeit widmet sich der Simulation und der Prognose des Materialverhaltens des Hochleistungsverbundwerkstoffes Textilbeton unter Zugbeanspruchungen. Basierend auf einer hierarchischen mechanischen Modellbildung (Multi-Skalen-Analyse) werden die Tragmechanismen des Verbundwerkstoffes auf drei Strukturebenen abgebildet. Damit lassen sich die den Verbundwerkstoff charakterisierenden mechanischen Kenngrößen aus experimentell ermittelten Kraft-Verschiebungs-Abhängigkeiten ableiten. Diese Kenngrößen sind mit heutiger Messtechnik nicht direkt experimentell bestimmbar. Es wird ein Mikro-Meso-Makro-Prognosemodell (MMM-Prognosemodell) für Textilbeton entwickelt, das basierend auf der Simulation des Mikrostrukturverhaltens das makroskopische Materialverhalten prognostiziert. Die Grundlage dafür bildet die qualitative und quantitative Bestimmung der Verbundeigenschaften zwischen der Filamentbewehrung und der einbettenden Matrix. Für das Verbundverhalten von Rovings in einer Feinbetonmatrix wird, ausgehend von einer Rovingapproximation mit superelliptischem Querschnitt, die partielle Imprägnierung des Rovings und die daraus resultierende Verbundwirkung identifiziert und simuliert. Auf Grundlage der mikro- und mesomechanischen Modelle sowie der Kalibrierung und Verifizierung des MMM-Prognosemodells durch die Simulation von Filament- und Rovingauszugsversuchen wird das makroskopische Zugverhalten von Textilbeton mit Mehrfachrissbildung prognostiziert. Die numerischen Ergebnisse werden durch die Ergebnisse der experimentellen Dehnkörperversuche validiert. Das MMM-Prognosemodell für Textilbeton wird im Rahmen einer hierarchischen Multi-Skalen-Analyse auf Zugversuche von Textilbetonbauteilen angewendet. Weiterhin wird die Verstärkungswirkung einer Textilbetonschicht an Stahlbetonbauteilen unter Biegebeanspruchung zutreffend simuliert. Es wird das nichtlineare Bauteilverhalten abgebildet, wobei die Bauteildurchbiegung, die effektiven Rovingbeanspruchungen und die Beanspruchungen der Filamente im Roving abgebildet werden.<br>The present work deals with the simulation and the prediction of the effective material behavior of the high performance composite textile reinforced concrete (TRC) subjected to tension. Based on a hierarchical material model within a multi scale approach the load bearing mechanisms of TRC are modeled on three structural scales. Therewith, the mechanical parameters characterizing the composite material can be deduced indirectly by experimentally determined force displacement relations obtained from roving pullout tests. These parameters cannot be obtained by contemporary measuring techniques directly. A micro-meso-macro-prediction model (MMM-PM) for TRC is developed, predicting the macroscopic material behavior by means of simulations of the microscopic and the mesoscopic material behavior. The basis is the qualitative and quantitative identification of the bond properties of the roving-matrix system. The partial impregnation of the rovings and the corresponding varying bond qualities are identified to characterize the bond behavior of rovings in a fine-grained concrete matrix. The huge variety of roving cross-sections is approximated by superellipses on the meso scale. The macroscopic behavior of TRC subjected to tension including multiple cracking of the matrix material is correctly predicted on the basis of the micro- and meso-mechanical models. The calibration and verification of the MMM-PM is performed by simulations of roving pullout tests, whereas a first validation is carried out by a comparison of the numerical predictions with the experimental data from tensile tests. The MMM-PM for TRC is applied to tensile tests of structural members made of TRC. Furthermore, a steel-reinforced concrete plate strengthened by a TRC layer is accurately simulated yielding the macroscopic deflection of the plate, the mesoscopic stress state of the roving and the microscopic stresses of the filaments.

This thesis, within the subfield of computer science known as computer vision, deals with the use of scale-space analysis in early low-level processing of visual information. The main contributions comprise the following five subjects: The formulation of a scale-space theory for discrete signals. Previously, the scale-space concept has been expressed for continuous signals only. We propose that the canonical way to construct a scale-space for discrete signals is by convolution with a kernel called the discrete analogue of the Gaussian kernel, or equivalently by solving a semi-discretized version of the diffusion equation. Both the one-dimensional and two-dimensional cases are covered. An extensive analysis of discrete smoothing kernels is carried out for one-dimensional signals and the discrete scale-space properties of the most common discretizations to the continuous theory are analysed. A representation, called the scale-space primal sketch, which gives a formal description of the hierarchical relations between structures at different levels of scale. It is aimed at making information in the scale-space representation explicit. We give a theory for its construction and an algorithm for computing it. A theory for extracting significant image structures and determining the scales of these structures from this representation in a solely bottom-up data-driven way. Examples demonstrating how such qualitative information extracted from the scale-space primal sketch can be used for guiding and simplifying other early visual processes. Applications are given to edge detection, histogram analysis and classification based on local features. Among other possible applications one can mention perceptual grouping, texture analysis, stereo matching, model matching and motion. A detailed theoretical analysis of the evolution properties of critical points and blobs in scale-space, comprising drift velocity estimates under scale-space smoothing, a classification of the possible types of generic events at bifurcation situations and estimates of how the number of local extrema in a signal can be expected to decrease as function of the scale parameter. For two-dimensional signals the generic bifurcation events are annihilations and creations of extremum-saddle point pairs. Interpreted in terms of blobs, these transitions correspond to annihilations, merges, splits and creations. Experiments on different types of real imagery demonstrate that the proposed theory gives perceptually intuitive results.<br><p>QC 20120119</p>

The paper presents a meso-scopic modeling framework for the simulation of three-phase composite consisting of a brittle cementitious matrix and reinforcing AR-glass yarns impregnated with epoxy resin. The construction of the model is closely related to the experimental program covering both the meso-scale test (yarn tensile test and double sided pull-out test) and the macro-scale test in the form of tensile test on the textile reinforced concrete specimen. The predictions obtained using the model are validated using a-posteriori performed experiments.

碩士<br>國立中央大學<br>資訊工程學系<br>105<br>In recent years, mobile device become more popular. And due to convenient transportation, people have higher probability to visit strange places. It is not easy to find a point of interest in a strange places, so we need to provide an electronic map system for users. It is not enough to provide name search for users only, because the users may not know the exact name of points. They may just want to find a specific category of point, so a good electronic map system needs to provide category search service. In order to provide category search services, we need to classify all the points in the system. Because the system has many points, each item has one or more categories, so this is a large-scale multi-label classification problem. There are many kind of categories, we follow the categories defined by Chinese yellow pages. The category consists two levels. There are 29 categories in level 1and 1,287 in level 2. Because the number of points and categories are large, we need to spend much time for training classifiers and testing data. We reduce the dimension of categories to speed up training and testing. After the experiment, our method’s training time and testing time are superior to the general SVM classification, the performance in level 1 Micro-F1 is 0.813, in level 2 Micro-F1 is 0.718 all slightly lower than SVM in level 1 Micro-F1 is 0.842. In level 2 Micro-F1 is 0.783. We want to try Reweighting, Downsampling to improve performance, but the performance is not wall in large-scale data.

碩士<br>國立臺灣科技大學<br>資訊工程系<br>105<br>For analyzing digital images of paintings we propose a new approach to categorize them based on artist. Determining the authorship of a painting is challenging because common subjects are illustrated in paintings, and paintings of an artist may not have a unique style. The proposed approach is built upon convolutional neural networks (CNN) a class of biologically inspired vision model that recently demonstrates near-human performance on several visual recognition tasks. However, training a CNN model requires large scale training data of a fixed input image size (e.g. 227 x 227). In this thesis, we propose to construct a multi-layer pyramid from an image, providing 21X more features than using a single layer (i.e., the original image) alone. We train a CNN model for each layer, and consider relationship within neighborhood patches in layers that have fine sub-regions, and propose a new weighted fusion scheme to aggregate the three layers adaptively. To evaluate the proposed methods, we collect two painting image datasets. The first dataset is categorized into 13 artists while the second one is categorized into 23 artists. As demonstrated in the experimental results, the proposed methods surpass the baseline method significantly in terms of precision, recall, and F-score to prove the effectiveness of proposed methods.

碩士<br>亞洲大學<br>資訊工程學系<br>104<br>Gender classification plays a significant role in many applications such as video surveillance and commercial advertisement. Human face shows us various information such as age, race, identity, gender, emotion and so on. We can obtain this facial information through various techniques such as human face tracking and detection, face recognition, gender classification, emotion recognition, age estimation and so on. In this thesis, we propose a new method of gender classification based on run-lengths histograms. The proposed method uses a run-length histogram to record the position information of pixels and efficiently improve the recognition rate, which is suitable for a big-data multimedia database. The experimental results show that the proposed method can achieve a better accuracy than that of multi-scale based method.

In recent years, multi-resolution techniques have become increasingly popular in the image processing community. New techniques have been developed with applications ranging from edge detection, texture recognition, image registration, multi-resolution features for image classification and more. The central focus of this two-part thesis is the multi-resolution analysis of images. In the first part, we used multi-resolution approaches to help with the classification of a set of protein crystal images. In the second, similar approaches were used to help register a set of 3D image volumes that would otherwise be computationally prohibitive without leveraging multi-resolution techniques. Specifically, the first part of this work proposes a classification framework that is being developed in collaboration with NorthEast Structural Genomics Consoritum (NESG) to assist in the automated screening of protein crystal images. Several groups have previously proposed automated algorithms to expedite such analysis. However, none of the classifiers described in the literature are sufficiently accurate or fast enough to be practical in a structural genomics production pipeline. The second part of this work proposes a 3D image registration algorithm to register regions of emphysema as quantified by densitometry on lung CT with MR lung volumes. The ability to register quantitatively-determined regions of emphysema with perfusion MRI will allow for further exploration of the pathophysiology of Chronic Obstructive Pulmonary Disorder (COPD). The registration method involves the registration of CT volumes at different levels of inspiration (total lung capacity to functional residual capacity [FRC]) followed by another registration between FRC-CT and FRC-MR volume pairs.

碩士<br>中華大學<br>土木工程學系碩士班<br>91<br>As the widespread of remote sensing application and the availability of various commercial satellites, there is an increasing number of satellite images could be adopted in research. Each satellite provides different image and scale characteristics. Generally speaking, large scale image provides wide range of information while small scale image provides local and detailed information. It should be able to find the most appropriate image scale for the purpose of different applications. The scale transfer is to convert one image scale to another in order to take advantage of the benefit from each image scale. The purpose of the study was to use the bilinear interpolation and the cubic convolution scale transferring techniques, converting high resolution Quick Bird image to 1, 2, 4, 10, 30, 100, and 250 m resolutions, respectively, and to examine the accuracy of the classified images, in order to find the most appropriate image scale for agricultural use in Taiwan. The results showed that the high resolution images after downscale transfer had not necessary providing more information. The pixels were just interpolated into a finer resolution. As a result, the disk space to hold the images increased exponentially. The up-scaled images tend to aggregate some of ground information, so that they seemed fuzzy and smooth. The overall accuracy of the paddy field land use classification indicated that the most appropriate image scale emerged at the size of land parcel. It showed no apparent difference between bilinear interpolation and cubic convolution methods. The results of the study indicated that it is not the higher the better for image scale. The scale should be in accordance with the ground characteristics. High resolution images usually denote high cost, smaller coverage, and longer processing time. According to the results of the study, it is not necessary to use very high resolution images for paddy field classification. A resolution approximately to the size of land parcel is the most appropriate since it reduces the cost while maintaining enough accuracy.

Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies.<br>This study investigates the potential of using very high resolution (VHR) QuickBird data to conduct vegetation classification of the Sundarban mangrove forest in Bangladesh and compares the results with Landsat TM data. Previous studies of vegetation classification in Sundarban involved Landsat images using pixel-based methods. In this study, both pixelbased and object-based methods were used and results were compared to suggest the preferred method that may be used in Sundarban. A hybrid object-based classification method was also developed to simplify the computationally demanding object-based classification, and to provide a greater flexibility during the classification process in absence of extensive ground validation data. The relation between NDVI (Normalized Difference Vegetation Index) and canopy cover was tested in the study area to develop a method to classify canopy cover type using NDVI value. The classification process was also designed with three levels of thematic details to see how different thematic scales affect the analysis results using data of different spatial resolutions. The results show that the classification accuracy using QuickBird data stays higher than that of Landsat TM data. The difference of classification accuracy between QuickBird and Landsat TM remains low when thematic details are low, but becomes progressively pronounced when thematic details are higher. However, at the highest level of thematic details, the classification was not possible to conduct due to a lack of appropriate ground validation data.(...)

碩士<br>淡江大學<br>電機工程學系碩士班<br>104<br>Visual tracking acts an important role in fields of computer vision as it can apply on robots to accomplish many tasks. In this thesis, a robust visual tracking algorithm is proposed based on the existing compressive tracking method. The proposed algorithm adopts multiple Bayes feature classifiers, each of them was trained under a different scale condition, to realize online multi-scale classification, which can greatly improve the robustness of tracking system. Furthermore, each feature classifier was initialized by randomly generating different types of Haar-like features. By doing so, the robustness of feature classification can be improved to obtain more accurate tracking results. In order to enhance real-time performance of visual tracking system, the processing time of each stage of the proposed algorithm was recorded. Then, the processing speed of the proposed algorithm is accelerated by adopting CPU-based parallel processing on the computationally expensive stages. Also, the formula of the Bayes feature classifier is simplified by numerical analysis to speedup the processing speed of multi-scale feature classification. After accelerating by parallel computation and formula simplification, the proposed visual tracking algorithm reaches about 45 images per second (45 fps) tracking performance in dealing with images of 642x352 pixels. Experimental results show that the proposed algorithm outperforms four state-of-the-art visual tracking methods in terms of success rate, tracking accuracy, and visual comparison.

Sustainable water resource management requires understanding how hydrologic processes are impacted by environmental management and land-use decisions across multiple spatial and temporal scales. A key concept in this respect is hydrologic ecosystem services (HES), which are the water related ‘goods’ produced by the environment which are valuable to humans. This dissertation assesses a range of topics concerning HES in the Vouga basin (north-central Portugal), and their connection with land-cover and land-use practices. Specifically, the relationship between changes in forest and agricultural land-cover and management practices, and associated changes in HES were examined using a range of statistical and modeling approaches. To quantify the effects of different agricultural scenarios on both HES and potential stakeholders, the ‘Soil and Water Assessment Tool’ (SWAT) was utilized, in conjunction with economic assessment methods. The first research section (Section 6) of the dissertation assesses the trends in streamflow quantity and yield in the Águeda watershed (a sub-basin of the Vouga) over a 75-yr period which coincided with large-scale afforestation of Pinus pinaster and (later) Eucalyptus globulus. Counter to the findings from meta-analysis studies of the effect of forest change on water availability, this study did not detect statistically significant trends in streamflow. By contrast, these findings support the view that there are prerequisite climatic, pedological, and eco-physiological watershed conditions that are necessary to observe hydrologic impacts at the watershed scale (which are not present in the Águeda watershed). By contrast, the significant changes which were detected are related to baseflow, which correspond with different periods of afforestation, and may be attributable to the promotion of soil water repellency under the mature pine and eucalypt stands. In the second research section (Section 7), an assessment is carried out on the hydrologic and nitrate dynamics at the whole basin scale, using the SWAT model. This assessment indicated that there is a high degree of variability in nitrate export from the different parts of the basin, with the highest rates coming from the lower (agriculturally dominated portion) of the basin. The main flow pathways for nitrate export were found to be leaching from agricultural land-cover types, which consistently had the highest export for all land-use and pathways. These findings indicate that the water bodies at the highest risk of nitrate pollution in the Vouga basin are the groundwater aquifers. The final research section (Section 8) utilizes the SWAT model to examine how reduced rates of fertilizer inputs would affect nitrate leaching, crop yields, and agricultural profitability in the lower Vouga basin. This research found that reduced rates of fertilization would reduce the amount of leached nitrate substantially, but that this would also lead to a large decrease in crop yield and profitability. A large difference in the inefficiency (i.e. crop production vs. nitrate export) between different HRUs was found, which could provide a focus for potential management action. This research strongly indicates that such actions may be needed to reduce the negative impacts of this pollution on the value of the groundwater aquifers, and to avoid associated costs which are otherwise passed on to local water users (e.g. through higher water treatment costs). The overall findings of the dissertation highlight the importance of the upper (forested) basin as a drinking water supply area, given the prevalence of nitrate pollution in the lower basin. However, the historic afforestation in the Vouga basin has resulted in a reduction in baseflow, which is negative from a drinking water supply perspective. Therefore, while the forested uplands are beneficial from water quality standpoint (compared to intensive agriculture), they also have altered flow patterns in a manner which will reduce available supply. The findings from the upper basin contrast sharply with the lower basin, where there are potentially large negative HES impacts due to current agricultural practices. These practices will primarily impact groundwater aquifers, and therefore the water quality within the lower basin receive little benefit from the relatively high-quality water from the upper basin. This highlights the importance of considering the interconnectivity of HES across spatial scales, which will depend on the specific site characteristics of the river basin.

Embryonic development depends on the precise coordination of cell fate specification, patterning and morphogenesis. Although great strides have been made in the molecular understanding of each of these processes, how their interplay governs the formation of complex tissues remains poorly understood. New techniques for experimental manipulation and image quantification enable the study of development in unprecedented detail, resulting in new hypotheses on the interactions between known components. By expressing these hypotheses in terms of rules and equations, computational modeling and simulation allows one to test their consistency against experimental data. However, new computational methods are required to represent and integrate the network of interactions between gene regulation, signaling and biomechanics that extend over the molecular, cellular and tissue scales. In this thesis, I present a framework that facilitates computational modeling of multiscale multicellular systems and apply it to investigate pancreatic development and the formation of vascular networks. This framework is based on the integration of discrete cell-based models with continuous models for intracellular regulation and intercellular signaling. Specifically, gene regulatory networks are represented by differential equations to analyze cell fate regulation; interactions and distributions of signaling molecules are modeled by reaction-diffusion systems to study pattern formation; and cell-cell interactions are represented in cell-based models to investigate morphogenetic processes. A cell-centered approach is adopted that facilitates the integration of processes across the scales and simultaneously constrains model complexity. The computational methods that are required for this modeling framework have been implemented in the software platform Morpheus. This modeling and simulation environment enables the development, execution and analysis of multi-scale models of multicellular systems. These models are represented in a new domain-specific markup language that separates the biological model from the computational methods and facilitates model storage and exchange. Together with a user-friendly graphical interface, Morpheus enables computational modeling of complex developmental processes without programming and thereby widens its accessibility for biologists. To demonstrate the applicability of the framework to problems in developmental biology, two case studies are presented that address different aspects of the interplay between cell fate specification, patterning and morphogenesis. In the first, I focus on the interplay between cell fate stability and intercellular signaling. Specifically, two studies are presented that investigate how mechanisms of cell-cell communication affect cell fate regulation and spatial patterning in the pancreatic epithelium. Using bifurcation analysis and simulations of spatially coupled differential equations, it is shown that intercellular communication results in a multistability of gene expression states that can explain the scattered spatial distribution and low cell type ratio of nascent islet cells. Moreover, model analysis shows that disruption of intercellular communication induces a transition between gene expression states that can explain observations of in vitro transdifferentiation from adult acinar cells into new islet cells. These results emphasize the role of the multicellular context in cell fate regulation during development and may be used to optimize protocols for cellular reprogramming. The second case study focuses on the feedback between patterning and morphogenesis in the context of the formation of vascular networks. Integrating a cell-based model of endothelial chemotaxis with a reaction-diffusion model representing signaling molecules and extracellular matrix, it is shown that vascular network patterns with realistic morphometry can arise when signaling factors are retained by cell-modified matrix molecules. Through the validation of this model using in vitro assays, quantitative estimates are obtained for kinetic parameters that, when used in quantitative model simulations, confirm the formation of vascular networks under measured biophysical conditions. These results demonstrate the key role of the extracellular matrix in providing spatial guidance cues, a fact that may be exploited to enhance vascularization of engineered tissues. Together, the modeling framework, software platform and case studies presented in this thesis demonstrate how cell-centered computational modeling of multi-scale and multicellular systems provide powerful tools to help disentangle the complex interplay between cell fate specification, patterning and morphogenesis during embryonic development.

Groundwater flow modelling provides an important quantitative instrument for addressing issues related to the quantity and quality of groundwater and the connected water resources. Consequently, groundwater flow models have been developed and used ubiquitously in science to deepen the understanding of subsurface processes and their drivers as well as management and planning tools. The present work investigates how numerical models can be linked to field investigations and public databases to quantitatively approach questions in the area of groundwater research. The primary goal is to develop new, efficient ways to overcome limitations of the individual hydrological concepts for solving specific hydrological problems and to increase the understanding of practical applicability of different methods. For this purpose, tailor-made approaches were developed for different study areas covering diverse spatial scales: the hydrology of a small mining lake, the riparian aquifer at the scale of a single meander as well as the aquifer systems of a large-scale river basin in China. The first part of the work deals with the physical and mathematical modelling of water constituents balance in a meromictic mining lake in Lusatia. The capability of using a rather simple mass-balance model based on a sufficient dataset of field data to evaluate lake stratification and lake-groundwater interaction were shown. In the second part, a transient numerical groundwater flow model was developed for the riparian aquifer of a stream meander and was calibrated by three different salt tracer tests. The model was used to proof the reliability of subsurface travel times derived from time series analysis and to give insights in the riparian zone dynamics during changing hydraulic gradients. The third part of the work describes the methodology to conduct risk assessment of groundwater contamination on the large catchment scale of the Songhua River in China. A comprehensive literature study was conducted to get an overview about measurement data on water quality data in China. A three-dimensional numerical flow and mass transport model was applied to access the flow and matter transport dynamics in the aquifer system of a sub-basin considering changing groundwater exploitation scenarios. Consequently, numerical groundwater modelling was combined with processed remote sensing and web mapping service data to overcome field data limitations and to derive groundwater vulnerability, groundwater hazard and groundwater risk maps for the entire Songhua River Basin. Summarizing, this doctoral thesis could develop new methods of combining field measurements, data assimilation and aggregation from various sources and groundwater modelling strategies and successfully apply these methods to find solutions on problems of multiple scales and across water systems.<br>Die Grundwassermodellierung stellt eine wichtige wissenschaftliche Methode zur quantitativen Analyse von Fragestellungen zum Schutz der Menge und Güte der Grundwasserressourcen sowie der angeschlossenen Wasserkörper dar. Dementsprechend werden Grundwassermodelle sowohl für Planungs- und Bewertungszwecke im Wasserressourcenmanagement als auch zur wissenschaftlichen Erforschung der Prozesse im Untergrund entwickelt und angewendet. Die vorliegende Arbeit untersucht in diesem Rahmen, wie numerische Modelle, Feldmessungen und Daten generiert aus Fernerkundungsdaten und Webplattformen systematisch verknüpft werden können, um Fragestellungen im Bereich der Grundwasserforschung quantitativ zu beantworten. Das Ziel der Arbeit ist es neue effiziente Abläufe zu entwickeln, die die Limitierung der einzelnen Methoden überwinden und diese auf deren Anwendbarkeit für die Lösung spezifischer hydrologischer Probleme zu analysieren. Zu diesem Zweck wurden in dieser Doktorarbeit fallspezifische Lösungen für verschiedene Untersuchungsgebiete entwickelt, die sowohl in der räumlichen Skale als auch in den zu untersuchenden hydrologischen Fragestellungen eine große Diversität aufweisen. Im ersten Teil der Arbeit wurde die Massenbilanz von Wasserinhaltsstoffen in einem meromiktischen Tagebaurestsee im Lausitzer Revier durch physikalische und mathematische Modellierungsmethoden untersucht. Dabei konnte gezeigt werden, dass auf Basis einer gewonnenen mehrjährigen Zeitreihe von Messdaten ein einfaches Massenbilanzmodell in der Lage ist, sowohl Seeschichtungs- als auch Grundwasseraustauschdynamiken quantitativ zu beschreiben. Der zweite Teil der Arbeit umfasst die Entwicklung eines transienten numerischen Grundwassermodells für den quartären Uferaquifer im Bereich eines Flussmäanders der Selke welches anhand von Daten aus mehreren Salztracertests kalibriert wurde. Das Modell wurde dafür verwendet die transienten Verweilzeiten in der gesättigten Zone des Mäanderbogens unter dem Einfluss dynamischer hydraulischer Bedingungen zu untersuchen. Die Ergebnisse wurden im Anschluss mit Verweilzeiten verglichen, die aus der Analyse der zeitlichen Verschiebung von gemessenen elektrischen Leitfähigkeitszeitreihen zwischen Fluss und Grundwassermessstellen gewonnen wurden. Durch dieses kombinierte Verfahren konnten sowohl die Beschränkungen der zeitreihenbasierten Verweilzeitberechnung aufgezeigt als auch ein tieferes Systemverständnis für die Interaktionsdynamiken zwischen Grund- und Flusswasser auf der Mäanderskala gewonnen werden. Der dritte Teil der Arbeit beschreibt die Vorgehensweise für die Bewertung des Grundwasserkontaminationsrisikos im Einzugsgebiet des Songhua Flusses in China. Eine umfassende Literaturstudie wurde durchgeführt, um einen Überblick über die Verfügbarkeit von Messdaten zur Belastung der Wasserressourcen Chinas mit organischen Schadstoffen zu erhalten. Danach wurde für ein Teileinzugsgebiet ein dreidimensionales numerisches Grundwassermodell auf Basis der vorhandenen hydrogeologischen Daten aufgebaut. Dieses wurde dazu verwendet die Änderungen im Stofftransports und den Schadstoffkonzentrationen innerhalb des Aquifersystems unter steigenden Entnahmeraten zu analysieren. Basierend auf diesen Studien wurden auf der Skale des Gesamteinzugsgebiets, um die beschränkte Verfügbarkeit von Felddaten auszugleichen, die Ergebnisse der numerischen Grundwassermodellierung mit Fernerkundungsdaten und Webdatenbanken in einem Indexsystem kombiniert mit dem für die oberflächennahen Aquifere Vulnerabilität, Gefährdungspotential und Verschmutzungsrisiko in einer räumlichen Auflösung von 1 km² bestimmt wurden. Zusammenfassend konnten durch die vorliegende Doktorarbeit neue passgenaue Methoden zur effektiven Kombination von in-situ Messungen, der Datenerhebung und Datenintegration aus vielfältigen Datenquellen sowie numerischen Grundwassermodellierungsstrategien entwickelt und zur Lösung der untersuchten hydrologischer Fragestellen auf den verschiedenen Skalen und über die Grenzen der einzelnen hydrologischen Teilsysteme hinaus erfolgreich angewandt werden.

The growing pressure on natural resources and biodiversity has led to a widespread acknowledgement of the services nature provides for humans. The appropriate representation of the benefits of sustainable land use in decision-making is still a challenge and tools that facilitate land use planning processes are needed. The dissertation presents a multicriteria assessment approach for a qualitative estimation of regional potentials to provide ecosystem services. This approach has been applied in several case studies in Saxony, Eastern Germany and Brasil. The ecosystem services concept builts the methodological framework for the assessment as it offers a universal approach to evaluate the impact of Land use/ Land cover change (LULCC) on human well-being. Since standardized methodical approaches for ecosystem services assessment at the landscape level are lacking, a particular requirement was to conceive a method that is easily transferable to other case study areas. Further the method should enable the use of existing and easily available environmental data, and it should be transparent for stakeholders and decision makers. The results of our study show that the combination of selected ecosystem services and land cover data such as CORINE Land Cover (CLC) can contribute to regional planning by communicating the effect of LULCC on ecosystem services, especially when applied as an evaluation basis in the tool GISCAME. The approach supports also the assessment of the performance of a region to provide ecosystem services and the comparison of regions towards this aspect. In the discussion section, the limitations of the developed approach are discussed. Main sources of uncertainty are related to coarse land cover data, lacking knowledge on the provision of ecosystem services at a landscape scale, and the difficulty to make relevant the ecosystem services concept in regional planning processes.