Dissertations / Theses on the topic 'Semi-automatic classification'

PONTIFÍCIA UNIVERSIDADE CATÓLICA DO RIO DE JANEIRO
O problema da classificação de dados remonta à criação de taxonomias visando cobrir áreas do conhecimento. Com o surgimento da Web, o volume de dados disponíveis aumentou várias ordens de magnitude, tornando praticamente impossível a organização de dados manualmente. Esta dissertação tem por objetivo organizar dados semi-estruturados, representados por frames, sem uma estrutura de classes prévia. A dissertação apresenta um algoritmo, baseado no K-Medóide, capaz de organizar um conjunto de frames em classes, estruturadas sob forma de uma hierarquia estrita. A classificação dos frames é feita a partir de um critério de proximidade que leva em conta os atributos e valores que cada frame possui.
The problem of data classification goes back to the definition of taxonomies covering knowledge areas. With the advent of the Web, the amount of data available has increased several orders of magnitude, making manual data classification impossible. This dissertation proposes a method to automatically classify semi-structured data, represented by frames, without any previous knowledge about structured classes. The dissertation introduces an algorithm, based on K-Medoid, capable of organizing a set of frames into classes, structured as a strict hierarchy. The classification of the frames is based on a closeness criterion that takes into account the attributes and their values in each frame.

A huge effort has been made in the development of image classification systemswith the objective of creating high-quality thematic maps and to establishprecise inventories about land cover use. The peculiarities of Remote SensingImages (RSIs) combined with the traditional image classification challengesmake RSI classification a hard task. Many of the problems are related to therepresentation scale of the data, and to both the size and therepresentativeness of used training set.In this work, we addressed four research issues in order to develop effectivesolutions for interactive classification of remote sensing images.The first research issue concerns the fact that image descriptorsproposed in the literature achieve good results in various applications, butmany of them have never been used in remote sensing classification tasks.We have tested twelve descriptors that encodespectral/color properties and seven texture descriptors. We have also proposeda methodology based on the K-Nearest Neighbor (KNN) classifier for evaluationof descriptors in classification context. Experiments demonstrate that JointAuto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition(SID), and Quantized Compound Change Histogram (QCCH) yield the best results incoffee and pasture recognition tasks.The second research issue refers to the problem of selecting the scaleof segmentation for object-based remote sensing classification. Recentlyproposed methods exploit features extracted from segmented objects to improvehigh-resolution image classification. However, the definition of the scale ofsegmentation is a challenging task. We have proposedtwo multiscale classification approaches based on boosting of weak classifiers.The first approach, Multiscale Classifier (MSC), builds a strongclassifier that combines features extracted from multiple scales ofsegmentation. The other, Hierarchical Multiscale Classifier (HMSC), exploits thehierarchical topology of segmented regions to improve training efficiencywithout accuracy loss when compared to the MSC. Experiments show that it isbetter to use multiple scales than use only one segmentation scale result. Wehave also analyzed and discussed about the correlation among the useddescriptors and the scales of segmentation.The third research issue concerns the selection of training examples and therefinement of classification results through multiscale segmentation. We have proposed an approach forinteractive multiscale classification of remote sensing images.It is an active learning strategy that allows the classification resultrefinement by the user along iterations. Experimentalresults show that the combination of scales produces better results thanisolated scales in a relevance feedback process. Furthermore, the interactivemethod achieves good results with few user interactions. The proposed methodneeds only a small portion of the training set to build classifiers that are asstrong as the ones generated by a supervised method that uses the whole availabletraining set.The fourth research issue refers to the problem of extracting features of ahierarchy of regions for multiscale classification. We have proposed a strategythat exploits the existing relationships among regions in a hierarchy. Thisapproach, called BoW-Propagation, exploits the bag-of-visual-word model topropagate features along multiple scales. We also extend this idea topropagate histogram-based global descriptors, the H-Propagation method. The proposedmethods speed up the feature extraction process and yield good results when compared with globallow-level extraction approaches.

Orientadores: Ricardo da Silva Torres, Alexandre Xavier Falcão
Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Computação
Made available in DSpace on 2018-08-23T15:18:27Z (GMT). No. of bitstreams: 1 Santos_JeferssonAlexdos_D.pdf: 18672412 bytes, checksum: 58ac60d8b5342ab705a78d5c82265ab8 (MD5) Previous issue date: 2013
Resumo: Um grande esforço tem sido feito para desenvolver sistemas de classificação de imagens capazes de criar mapas temáticos de alta qualidade e estabelecer inventários precisos sobre o uso do solo. As peculiaridades das imagens de sensoriamento remoto (ISR), combinados com os desafios tradicionais de classificação de imagens, tornam a classificação de ISRs uma tarefa difícil. Grande parte dos desafios de pesquisa estão relacionados à escala de representação dos dados e, ao mesmo tempo, à dimensão e à representatividade do conjunto de treinamento utilizado. O principal foco desse trabalho está nos problemas relacionados à representação dos dados e à extração de características. O objetivo é desenvolver soluções efetivas para classificação interativa de imagens de sensoriamento remoto. Esse objetivo foi alcançado a partir do desenvolvimento de quatro linhas de pesquisa. A primeira linha de pesquisa está relacionada ao fato de embora descritores de imagens propostos na literatura obterem bons resultados em várias aplicações, muitos deles nunca foram usados para classificação de imagens de sensoriamento remoto. Nessa tese, foram testados doze descritores que codificam propriedades espectrais e sete descritores de textura. Também foi proposta uma metodologia baseada no classificador K-Vizinhos mais Próximos (K-nearest neighbors - KNN) para avaliação de descritores no contexto de classificação. Os descritores Joint Auto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition (SID) e Quantized Compound Change Histogram (QCCH), apresentaram os melhores resultados experimentais na identificação de alvos de café e pastagem. A segunda linha de pesquisa se refere ao problema de seleção de escalas de segmentação para classificação de imagens de sensoriamento baseada em objetos. Métodos propostos recentemente exploram características extraídas de objetos segmentados para melhorar a classificação de imagens de alta resolução. Entretanto, definir uma escala de segmentação adequada é uma tarefa desafiadora. Nessa tese, foram propostas duas abordagens de classificação multiescala baseadas no algoritmo Adaboost. A primeira abordagem, Multiscale Classifier (MSC), constrói um classificador forte que combina características extraídas de múltiplas escalas de segmentação. A outra, Hierarchical Multiscale Classifier (HMSC), explora a relação hierárquica das regiões segmentadas para melhorar a eficiência sem reduzir a qualidade da classificação xi quando comparada à abordagem MSC. Os experimentos realizados mostram que é melhor usar múltiplas escalas do que utilizar apenas uma escala de segmentação. A correlação entre os descritores e as escalas de segmentação também é analisada e discutida. A terceira linha de pesquisa trata da seleção de amostras de treinamento e do refinamento dos resultados da classificação utilizando segmentação multiescala. Para isso, foi proposto um método interativo para classificação multiescala de imagens de sensoriamento remoto. Esse método utiliza uma estratégia baseada em aprendizado ativo que permite o refinamento dos resultados de classificação pelo usuário ao longo de interações. Os resultados experimentais mostraram que a combinação de escalas produzem melhores resultados do que a utilização de escalas isoladas em um processo de realimentação de relevância. Além disso, o método interativo obtém bons resultados com poucas interações. O método proposto necessita apenas de uma pequena porção do conjunto de treinamento para construir classificadores tão fortes quanto os gerados por um método supervisionado utilizando todo o conjunto de treinamento disponível. A quarta linha de pesquisa se refere à extração de características de uma hierarquia de regiões para classificação multiescala. Assim, foi proposta uma abordagem que explora as relações existentes entre as regiões da hierarquia. Essa abordagem, chamada BoW-Propagation, utiliza o modelo bag-of-visual-word para propagar características ao longo de múltiplas escalas. Essa ideia foi estendida para propagar descritores globais baseados em histogramas, a abordagem H-Propagation. As abordagens propostas aceleram o processo de extração e obtém bons resultados quando comparadas a descritores globais
Abstract: A huge effort has been made in the development of image classification systems with the objective of creating high-quality thematic maps and to establish precise inventories about land cover use. The peculiarities of Remote Sensing Images (RSIs) combined with the traditional image classification challenges make RSI classification a hard task. Many of the problems are related to the representation scale of the data, and to both the size and the representativeness of used training set. In this work, we addressed four research issues in order to develop effective solutions for interactive classification of remote sensing images. The first research issue concerns the fact that image descriptors proposed in the literature achieve good results in various applications, but many of them have never been used in remote sensing classification tasks. We have tested twelve descriptors that encode spectral/color properties and seven texture descriptors. We have also proposed a methodology based on the K-Nearest Neighbor (KNN) classifier for evaluation of descriptors in classification context. Experiments demonstrate that Joint Auto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition (SID), and Quantized Compound Change Histogram (QCCH) yield the best results in coffee and pasture recognition tasks. The second research issue refers to the problem of selecting the scale of segmentation for object-based remote sensing classification. Recently proposed methods exploit features extracted from segmented objects to improve high-resolution image classification. However, the definition of the scale of segmentation is a challenging task. We have proposed two multiscale classification approaches based on boosting of weak classifiers. The first approach, Multiscale Classifier (MSC), builds a strong classifier that combines features extracted from multiple scales of segmentation. The other, Hierarchical Multiscale Classifier (HMSC), exploits the hierarchical topology of segmented regions to improve training efficiency without accuracy loss when compared to the MSC. Experiments show that it is better to use multiple scales than use only one segmentation scale result. We have also analyzed and discussed about the correlation among the used descriptors and the scales of segmentation. The third research issue concerns the selection of training examples and the refinement of classification results through multiscale segmentation. We have proposed an approach for xix interactive multiscale classification of remote sensing images. It is an active learning strategy that allows the classification result refinement by the user along iterations. Experimental results show that the combination of scales produces better results than isolated scales in a relevance feedback process. Furthermore, the interactive method achieves good results with few user interactions. The proposed method needs only a small portion of the training set to build classifiers that are as strong as the ones generated by a supervised method that uses the whole available training set. The fourth research issue refers to the problem of extracting features of a hierarchy of regions for multiscale classification. We have proposed a strategy that exploits the existing relationships among regions in a hierarchy. This approach, called BoW-Propagation, exploits the bag-of-visual-word model to propagate features along multiple scales. We also extend this idea to propagate histogram-based global descriptors, the H-Propagation method. The proposed methods speed up the feature extraction process and yield good results when compared with global low-level extraction approaches
Doutorado
Ciência da Computação
Doutor em Ciência da Computação

Cette thèse présente un chaine d'analyse d'image qui, à partir d'images numériques à haute résolution et à trois ou quatre canaux (50 cm, couleur et, dans certains cas, proche infrarouge), mais aussi en s'appuyant sur le parcellaire cadastral, rend une segmentation des images en parcelles agraires (champs, forêts, vignes,. . . ) et une classification de celles-ci, avec une très haute fiabilité, et attribue à chaque segment classifié une mesure qui indique la confiance que le système a en cette classification. Elle inclut une étude sur l'intérêt de la texture et les espaces de couleur pour la segmentation et la classification, deux méthodes de recalage de graphes sur une image, un modèle de probabilité novateur et ses algorthmes de classification par régions associées, et un éstimateur très précis de la période et de l'orientation
This thesis presents a complete image analisys system which, from high-resolution 3 or 4-channel digital images (50 cm, colour and optionally near infrared), and using the cadastre database, segments the images into agriculturally-homogeneous regions, (fields, forests, vines, and so on) and classifies these regions, tagging each classified region with a confidence measure which indicates the system's confidence in each classification. It includes a study of the value of texture features and transformed colour spaces for segmentation and classification, two methods for registering a graph onto an image, a novel probability model and associated per-region classification algorithms, and a high precision period and orientation estimator

As recentes tecnologias de desenvolvimento e distribuição de componentes possibilitaram o aumento do número de componentes disponíveis no mercado. No entanto, eles muitas vezes estão dispersos e não publicados adequadamente para a comunidade de pesquisa e desenvolvimento de software. Encontrar componentes apropriados para solucionar um problema particular não é uma tarefa simples e novas técnicas devem ser desenvolvidas para o reuso efetivo de componentes. Um dos maiores desafios em reusar componentes consiste em classificá-los corretamente para futuras consultas. Classificar componentes para possibilitar uma busca eficaz depende da qualidade das informações adquiridas, que viabilizam melhor precisão e cobertura das consultas ao encontrar componentes reutilizáveis em potencial. Ao mesmo tempo, mecanismos de classificação e busca devem ser fáceis o suficiente para convencer os desenvolvedores a reusar componentes. Este trabalho estuda as técnicas de classificação de componentes de software, repositórios e métodos de busca. é apresentada uma proposta de modelo de classificação de componentes que considera não apenas sua função, mas o negócio onde ele está inserido e seus atributos de qualidade. Um método de preenchimento semi-automático das informações é proposto, de modo a diminuir os custos de classificação. O protótipo REUSE+ foi construído para exemplificar o uso do modelo e do método de classificação semi-automática, de forma a validar a proposta, destacando, por fim, as principais contribuições do trabalho.
The recent developments on components technologies have increased the number of components available to the market. These components are, however, distributed overall the world and not properly advertised to the research and development communities. Finding the appropriate components to solve a particular problem is not very straightforward and new techniques must be developed to effectively reuse components. One of the great challenges in reusing components is concerned with how to actually classify components \"properly\" in order to further retrieve them. Classifying components for effective retrieval depends on acquiring the appropriate information in classification to improve the precision and recall rates in retrieval; finding only the potentially reusable components and not missing potential solutions. At the same time, the classification and retrieval mechanisms must be easy enough to persuade developers to reuse components. This work studies the classification techniques of software components, repository and retrieval methods. Hereafter is presented a proposal of components classification model that considers not just its function, but business and quality attributes. It is proposed a semi-automatic classification mechanism of software information, allowing a cheaper classification. REUSE+ prototype was built to exemplify the use of model and method of semi-automatic classification, allowing the described proposal validation, highlighting at the end the mainly contributions of the work.

Includes abstract.
Includes bibliographical references.
It is essential for geospatial and mapping organisations that changes to the landscapeare regularly detected and captured, so that map databases can be updated. The Chief Directorate of National Geospatial Information (CD: NGI), South Africa’s national mapping agency, currently relies on manual methods for digitizing features and detecting changes. These methods are time consuming and labour intensive, and rely on the skills and interpretation of the operator. It is therefore necessary to move towards more automated methods in the production process at CD: NGI. The objective of this research is to develop a process for semi-automatic classification of built-up areas from aerial imagery in South Africa. Built-up areas are important as they can grow and change rapidly. Since the South African landscape is varied and climatological conditions differ from one area to another, a general and robust method that can be applied across the country is needed. This project aims to find the best approach for classifying urban built-up areas from high-resolution aerial imagery by comparing various image classification methods, so that a method that is transferable and applicable in diverse South African scenes may be developed. Image classification methods were compared and it was found that pixel-based classifiers were unsatisfactory in classifying built-up areas, whereas object-based classifiers had better results. Image segmentation, the first step in an object-based classification, can considerably influence the results of the classification task. It is therefore essential that suitable image segments be generated before the segments are classified. The proposed The proposed methodology involves the use of cadastral data in the image segmentation process and texture measures in the classification of built-up areas within an object-based process. The method can be applied to diverse scenes across South Africa to find built-up areas. This is a generalised approach and can assist the CD: NGI in the process of updating their topographic database by reducing the time that operators spend on identifying and manually digitizing built-up areas.

The knowledge available to be exploited by text classification and information retrieval systems has significantly changed, both in nature and quantity, in the last years. Nowadays, there are several sources of information that can potentially improve the classification process, and systems should be able to adapt to incorporate multiple sources of available data in different formats. This fact is specially important in environments where the required information changes rapidly, and its utility may be contingent on timely implementation. For these reasons, the importance of adaptability and flexibility in information systems is rapidly growing. Current systems are usually developed for specific scenarios. As a result, significant engineering effort is needed to adapt them when new knowledge appears or there are changes in the information needs. This research investigates the usage of knowledge within text classification from two different perspectives. On one hand, the application of descriptive approaches for the seamless modelling of text classification, focusing on knowledge integration and complex data representation. The main goal is to achieve a scalable and efficient approach for rapid prototyping for Text Classification that can incorporate different sources and types of knowledge, and to minimise the gap between the mathematical definition and the modelling of a solution. On the other hand, the improvement of different steps of the classification process where knowledge exploitation has traditionally not been applied. In particular, this thesis introduces two classification sub-tasks, namely Semi-Automatic Text Classification (SATC) and Document Performance Prediction (DPP), and several methods to address them. SATC focuses on selecting the documents that are more likely to be wrongly assigned by the system to be manually classified, while automatically labelling the rest. Document performance prediction estimates the classification quality that will be achieved for a document, given a classifier. In addition, we also propose a family of evaluation metrics to measure degrees of misclassification, and an adaptive variation of k-NN.

PONTIFÍCIA UNIVERSIDADE CATÓLICA DO RIO DE JANEIRO
COORDENAÇÃO DE APERFEIÇOAMENTO DO PESSOAL DE ENSINO SUPERIOR
PROGRAMA DE SUPORTE À PÓS-GRADUAÇÃO DE INSTS. DE ENSINO
Os processos de transformação da paisagem são resultantes da interação de elementos (bióticos e abióticos) que compõe a superfície da Terra. Baseia-se, a partir de uma perspectiva holística, no inter-relacionamento de uma série de ações e objetos que confluem para que a paisagem seja percebida como um momento sintético da confluência de inúmeras temporalidades. Desta maneira, as geotecnologias passam a se constituir como um importante aparato técnico-científico para a interpretação desta realidade ao possibilitar novas e diferentes formas do ser humano interpretar a paisagem. Um dos produtos gerados a partir desta interpretação é a classificação de uso e cobertura do solo e que se configura como um instrumento central para a análise das dinâmicas territoriais. Desta maneira, o objetivo do presente trabalho é elaboração de um modelo de classificação semi-automática baseada em conhecimento geográfico para o levantamento do padrão de uso e cobertura da paisagem a partir da utilização de imagens de satélite de alta resolução, tendo como recorte analítico uma área na porção setentrional no Maciço da Tijuca. O modelo baseado na análise de imagens baseadas em objetos, quando confrontados com a classificação visual, culminou em um valor acima de 80 por cento de correspondência tanto para imagens de 2010 e 2009, apresentando valores bastante elevados também na comparação classe a classe. A elaboração do presente modelo contribuiu diretamente para a otimização da produção dos dados elaborados contribuindo sobremaneira para a aceleração da interpretação das imagens analisadas, assim como para a minimização de erros ocasionados pela subjetividade atrelada ao próprio classificador.
The transformation processes of the landscape are results from the interaction of factors (biotic and abiotic) that makes up the Earth s surface. This interaction, from a holistic perspective, is then based on the inter-relationship of a series of actions and objects that converge so that landscape is perceived as a moment of confluence of numerous synthetic temporalities. Thus, the geotechnologies come to constitute an important technical and scientific apparatus for the interpretation of this reality by enabling new and different ways of interpreting the human landscape. One of the products that can be generated from this interpretation is the use classification and land cover and is configured as a central instrument for the analysis of territorial dynamics. Thus, the aim of this work is the development of a semi-automatic classification model based on geographic knowledge to survey the pattern of land use and cover the landscape from the use of satellite images of high resolution, with the analytical approach an area in the northern portion of the Tijuca Massif. The model built on an Object-Based Image Analysis, when confronted with the visual classification, culminated in a value above 80 percent match for 2010 and 2009, with very high values in the comparison class to class. The development of this model directly contributed to the optimization of the production of processed data contributing greatly to the acceleration of the interpretation of the images analyzed, as well as to minimize errors caused by the subjectivity linked to the classifier itself.

From a global point of view, as argued by Levy (1999), the modern city has undergone radical changes in its physical form, either in terms of territorial expansion as well as in terms of interna! physical transformations. Today, approximately 75% of the European population lives in urban areas ,which makes the urban fulure of the conlinent a major cause of concern (Brazil, Cavalcanti, & Longo, 2014). lndeed, the demand for urban land, both within and around the cities, is becoming increasingly acule (European Environmenl Agency, 2006). Ouring the last decades, also Spain has been undergoing an important process of urban growth, which has implied the consumption of a large amounl of land, al hough the overall population growth rale, mostly along certain specific geographic areas, has remained at least unchanged or even, in sorne cases, il has also decreased. Such a phenomenon has been quite remarkable along the Mediterranean side. As argued by Gaja (2008), the urban development in Spain has been strongly linked to the model of economic development , which relies, since its launch in the 50's, onlhree main factors , i.e.:emigration, building, and mass tourism. Nowadays , in Spain, and mostly along the Medilerranean side, several urban areas are facing important phenomena of urban sprawl, also feared by he European Union. An accurate information about the pattern of land use/land cover, over time, is a fundamental requirement for a better understanding of the urban models. Currently, even though plenty of approaches to the image classification, through Remote Sensing (RS) techniques, have been advanced, Land Cover/Land Use classification is still an exciting challenge (Weng, 2010). Actually, the increasing development of RS and GIS technologies, during the last decades, has provided further capabiliies for measuring, analysing, understanding, modelling the "physical expressions" of urban growth phenomena, either in terms of pattern and process (Bhatta, 2012), and based on land use/land cover mapping and change delection over time. Based on such a technological approach, here we first aim to set up a suitable methodology for detecting generalized land cover classes based on an assisted automatic (or semi-aulomatic) pixel-based approach, calibrated upon Landsat Thematic Mapper (TM) mullispectral imagery, at 30 meters of spatial resolution. Beside, through the use of Geographical lnformation Syslem (GIS) we provide a spatial analysis and modelling of different urban models, from a morphological standpoint, in order to define the main pattern of land occupation al municipal scale, and along the Mediterranean side of Spain, al the year 2011. We focus on two main issues. On one hand, RS techniques have been used to set up a proper semi-automatic classification methodology, based on the use of Landsat imagery, capable of handling huge geographical areas quickly and efficiently. This process is basically aimed to detect the urban areas, at the year 2011, along the Mediterranean side of Spain, depending on the administrative division of Autonomous Communities. On the other hand, the spatial patterns of urban settlements have been analysed by using a GIS platform for quantifying a set of spatial metrics about the urban form. Hence, once get the quantification of different morphological features, including the analysis aboul either the urban profile, the urban texture, and the street network pattern, an automatic classification of different urban morphological models has been proposed, based on a stalistical approaches, namely factor and cluster analysis
Desde un punto de vista global,como sostiene Levy (1999), la ciudad moderna ha experimentado cambios radicales en su forma física, ya sea en términos de expansión territorial, así como en términos de transformaci ones internas. Hoy en día, aproximadamente el 75% de la población europea vive en zonas urbanas, lo que hace del futuro urbano delcontinente, una causa importante de preocupación (Brasil, Cavalcanti, y Longo, 2014). De hecho, la demanda de suelo urbano, dentro y alrededor de las ciudades , es cada vez más aguda (Agencia Europea de Medio Ambiente,2006). Durante las últimas décadas, también España ha experimentado un importante proceso de crecimiento urbano que ha implicado el consumo de una gran cantidad de tierra, aunque la tasa de crecimiento de la población en general, sobre todo a lo largo de ciertas áreas geográficas específicas , se ha mantenido al menos sin cambios o incluso, en algunos casos, también ha disminuido. Este fenómeno ha sido muy evidente a lo largo de la vertiente mediterránea. Como sostiene Gaja (2008), el desarrollo urbano en España se ha visto fuertemente vinculado con el modelo de desarrollo económico, que se basa, desde su lanzamiento en la década de los 50,en tres factores principales, a saber: la emigración, la construcción y el turismo de masas. Hoy en día, en España, y sobre todo a lo largo de la vertiente mediterránea, varias zonas urbanas se enfrentan a fenómenos importantes de expansión urbana, también temidos por la Unión Europea. Al respecto,un requisito fundamental para mejorar la comprensión y el estudio de los modelos urbanos es obtener en eltiempo una información precisa sobre los patrones de cubiertas y uso de suelo. Actualmente, a pesar de la existencia de numerosos métodos para la clasificación de imágenes digitales a través de técnicas de teledetección, para ext raer información sobre cobertura/uso de suelo, este enfoque sigue siendo un reto apasionante (Weng, 2010). El creciente desarrollo de las tecnologías de RS y GIS, durante las últimas décadas, ha proporcionado nuevas capacidades para medir, analizar, comprender, y modelar las "expresiones físicas" de los fenómenos de crecimiento urbano, en términos de patrones y procesos (Bhatta, 2012), y con base en el mapeo y análisis de cambios de cobertura/uso de suelo a través el tiempo. Basándose en un enfoque tecnológico, el primero objetivo es establecer una metodología adecuada para la detección de clases de cobertura de la tierra generalizadas que encuentra su fundamento en una asistido automático (o semiautomático), enfoque basado en píxeles, calibradas en Landsat Thematic Mapper (TM) imágenes multiespectrales, a 30 metros de resolución espacial. Al lado, a través del uso del Sistema de Información Geográfica (SIG), es posible proveer un análisis espacial y la modelización de diferentes modelos urbanos, desde un punto de vista morfológico, con el fin de definir el patrón principal de la ocupación del suelo a escala municipal a lo largo de la vertiente mediterránea de España, en el año 2011. En particular no enfocamos en dos cuestiones principales. Por un lado, las técnicas de RS se han utilizado para establecer una metodología de clasificación semi-automático adecuada, basada en el uso de imágenes Landsat, capaz de manejar grandes zonas geográficas de forma rápida y eficiente. Este proceso, básicamente, va dirigido a detectar las áreas urbanas, en el año 2011, a lo largo de la vertiente mediterránea de España, según la división administrativa de las Comunidades Autónomas. Por otro lado, los patrones espaciales de asentamientos urbanos han sido analizados mediante el uso de una plataforma GIS para cuantificar un conjunto de métricas espaciales sobre la forma urbana. Finalmente, una vez obtenida la cuantificación de diferentes características morfológicas, se ha proporcionado una clasificación automática de los diferentes modelos morfológicos urbanos, basada en un enfoque estadístico, es decir, análisis factorial y clúster.

In this thesis, we have studied the problem of detecting automated Twitter accounts related to the Ukraine conflict using supervised learning. A striking problem with the collected data set is that it was initially lacking a ground truth. Traditionally, supervised learning approaches rely on manual annotation of training sets, but it incurs tedious work and becomes expensive for large and constantly changing collections. We present a novel approach to synthetically generate large amounts of labeled Twitter accounts for detection of automation using a rule-based classifier. It significantly reduces the effort and resources needed and speeds up the process of adapting classifiers to changes in the Twitter-domain. The classifiers were evaluated on a manually annotated test set of 1,000 Twitter accounts. The results show that rule-based classifier by itself achieves a precision of 94.6% and a recall of 52.9%. Furthermore, the results showed that classifiers based on supervised learning could learn from the synthetically generated labels. At best, the these machine learning based classifiers achieved a slightly lower precision of 94.1% compared to the rule-based classifier, but at a significantly better recall of 93.9%
Detta exjobb har undersökt problemet att detektera automatiserade Twitter-konton relaterade till Ukraina-konflikten genom att använda övervakade maskininlärningsmetoder. Ett slående problem med den insamlade datamängden var avsaknaden av träningsexempel. I övervakad maskininlärning brukar man traditionellt manuellt märka upp en träningsmängd. Detta medför dock långtråkigt arbete samt att det blir dyrt förstora och ständigt föränderliga datamängder. Vi presenterar en ny metod för att syntetiskt generera uppmärkt Twitter-data (klassifieringsetiketter) för detektering av automatiserade konton med en regel-baseradeklassificerare. Metoden medför en signifikant minskning av resurser och anstränging samt snabbar upp processen att anpassa klassificerare till förändringar i Twitter-domänen. En utvärdering av klassificerare utfördes på en manuellt uppmärkt testmängd bestående av 1,000 Twitter-konton. Resultaten visar att den regelbaserade klassificeraren på egen hand uppnår en precision på 94.6% och en recall på 52.9%. Vidare påvisar resultaten att klassificerare baserat på övervakad maskininlärning kunde lära sig från syntetiskt uppmärkt data. I bästa fall uppnår dessa maskininlärningsbaserade klassificerare en något lägre precision på 94.1%, jämfört med den regelbaserade klassificeraren, men med en betydligt bättre recall på 93.9%.

[ES] Las imágenes de fondo de ojo son muy utilizadas por los oftalmólogos para la evaluación de la retina y la detección de glaucoma. Esta patología es la segunda causa de ceguera en el mundo, según estudios de la Organización Mundial de la Salud (OMS). En esta tesis doctoral, se estudian algoritmos de aprendizaje automático (machine learning) para la evaluación automática del glaucoma usando imágenes de fondo de ojo. En primer lugar, se proponen dos métodos para la segmentación automática. El primer método utiliza la transformación Watershed Estocástica para segmentar la copa óptica y posteriormente medir características clínicas como la relación Copa/Disco y la regla ISNT. El segundo método es una arquitectura U-Net que se usa específicamente para la segmentación del disco óptico y la copa óptica. A continuación, se presentan sistemas automáticos de evaluación del glaucoma basados en redes neuronales convolucionales (CNN por sus siglas en inglés). En este enfoque se utilizan diferentes modelos entrenados en ImageNet como clasificadores automáticos de glaucoma, usando fine-tuning. Esta nueva técnica permite detectar el glaucoma sin segmentación previa o extracción de características. Además, este enfoque presenta una mejora considerable del rendimiento comparado con otros trabajos del estado del arte. En tercer lugar, dada la dificultad de obtener grandes cantidades de imágenes etiquetadas (glaucoma/no glaucoma), esta tesis también aborda el problema de la síntesis de imágenes de la retina. En concreto se analizaron dos arquitecturas diferentes para la síntesis de imágenes, las arquitecturas Variational Autoencoder (VAE) y la Generative Adversarial Networks (GAN). Con estas arquitecturas se generaron imágenes sintéticas que se analizaron cualitativa y cuantitativamente, obteniendo un rendimiento similar a otros trabajos en la literatura. Finalmente, en esta tesis se plantea la utilización de un tipo de GAN (DCGAN) como alternativa a los sistemas automáticos de evaluación del glaucoma presentados anteriormente. Para alcanzar este objetivo se implementó un algoritmo de aprendizaje semi-supervisado.
[CAT] Les imatges de fons d'ull són molt utilitzades pels oftalmòlegs per a l'avaluació de la retina i la detecció de glaucoma. Aquesta patologia és la segona causa de ceguesa al món, segons estudis de l'Organització Mundial de la Salut (OMS). En aquesta tesi doctoral, s'estudien algoritmes d'aprenentatge automàtic (machine learning) per a l'avaluació automàtica del glaucoma usant imatges de fons d'ull. En primer lloc, es proposen dos mètodes per a la segmentació automàtica. El primer mètode utilitza la transformació Watershed Estocàstica per segmentar la copa òptica i després mesurar característiques clíniques com la relació Copa / Disc i la regla ISNT. El segon mètode és una arquitectura U-Net que s'usa específicament per a la segmentació del disc òptic i la copa òptica. A continuació, es presenten sistemes automàtics d'avaluació del glaucoma basats en xarxes neuronals convolucionals (CNN per les sigles en anglès). En aquest enfocament s'utilitzen diferents models entrenats en ImageNet com classificadors automàtics de glaucoma, usant fine-tuning. Aquesta nova tècnica permet detectar el glaucoma sense segmentació prèvia o extracció de característiques. A més, aquest enfocament presenta una millora considerable del rendiment comparat amb altres treballs de l'estat de l'art. En tercer lloc, donada la dificultat d'obtenir grans quantitats d'imatges etiquetades (glaucoma / no glaucoma), aquesta tesi també aborda el problema de la síntesi d'imatges de la retina. En concret es van analitzar dues arquitectures diferents per a la síntesi d'imatges, les arquitectures Variational Autoencoder (VAE) i la Generative adversarial Networks (GAN). Amb aquestes arquitectures es van generar imatges sintètiques que es van analitzar qualitativament i quantitativament, obtenint un rendiment similar a altres treballs a la literatura. Finalment, en aquesta tesi es planteja la utilització d'un tipus de GAN (DCGAN) com a alternativa als sistemes automàtics d'avaluació del glaucoma presentats anteriorment. Per assolir aquest objectiu es va implementar un algoritme d'aprenentatge semi-supervisat.
[EN] Fundus images are widely used by ophthalmologists to assess the retina and detect glaucoma, which is, according to studies from the World Health Organization (WHO), the second cause of blindness worldwide. In this thesis, machine learning algorithms for automatic glaucoma assessment using fundus images are studied. First, two methods for automatic segmentation are proposed. The first method uses the Stochastic Watershed transformation to segment the optic cup and measures clinical features such as the Cup/Disc ratio and ISNT rule. The second method is a U-Net architecture focused on the optic disc and optic cup segmentation task. Secondly, automated glaucoma assessment systems using convolutional neural networks (CNNs) are presented. In this approach, different ImageNet-trained models are fine-tuned and used as automatic glaucoma classifiers. These new techniques allow detecting glaucoma without previous segmentation or feature extraction. Moreover, it improves the performance of other state-of-art works. Thirdly, given the difficulty of getting large amounts of glaucoma-labelled images, this thesis addresses the problem of retinal image synthesis. Two different architectures for image synthesis, the Variational Autoencoder (VAE) and Generative Adversarial Networks (GAN) architectures, were analysed. Using these models, synthetic images that were qualitative and quantitative analysed, reporting state-of-the-art performance, were generated. Finally, an adversarial model is used to create an alternative automatic glaucoma assessment system. In this part, a semi-supervised learning algorithm was implemented to reach this goal.
The research derived from this doctoral thesis has been supported by the Generalitat Valenciana under the scholarship Santiago Grisolía [GRISOLIA/2015/027].
Díaz Pinto, AY. (2019). Machine Learning for Glaucoma Assessment using Fundus Images [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/124351
TESIS

Cette thèse porte sur le traitement automatique d'images couleur, et son application à la robotique dans des environnements semi-structurés d'extérieur. Nous proposons une méthode de navigation visuelle pour des robots mobiles en utilisant une caméra couleur. Les domaines d'application de ce travail se situent dans l'automatisation de machines agricoles, en vue de la navigation automatique dans un réseau de chemins (pour aller d'une ferme à un champ par exemple). Nous présentons tout d'abord une analyse des principaux travaux de recherche dans la littérature sur la navigation visuelle. Une chaîne de pré-traitement pour le rendu couleur d'images numériques mono-capteur dotées d'un filtre Bayer est présentée ; elle se base sur une étude des techniques de démosaïquage, le calibrage chromatique d'images (balance de blancs) et la correction gamma. Une méthode d'interprétation monoculaire de la scène courante permet d'extraire les régions navigables et un modèle 2D de la scène. Nous traitons de la segmentation d'une image couleur en régions, puis de la caractérisation de ces régions par des attributs de texture et de couleur, et enfin, de l'identification des diverses entités de la scène courante (chemin, herbe, arbre, ciel, champ labouré,&). Pour cela, nous exploitons deux méthodes de classification supervisée : la méthode de Support Vector Machine) (SVM) et celle des k plus proches voisins (k-PPV). Une réduction d'information redondante par une analyse en composantes indépendantes (ACI) a permis d'améliorer le taux global de reconnaissance. Dans un réseau de chemins, le robot doit reconnaître les intersections de chemins lui permettant (a) dans une phase d'apprentissage, de construire un modèle topologique du réseau dans lequel il va devoir se déplacer et (b) dans une phase de navigation, de planifier et exécuter une trajectoire topologique définie dans ce réseau. Nous proposons donc une méthode de détection et classification du chemin: ligne droite, virage gauch e, virage droite, carrefour en X, en T ou en Y. Une approche pour la représentation de la forme et de la catégorisation des contours (Shape Context) est utilisée à cet effet. Une validation a été effectuée sur une base d'images de routes ou chemins de campagne. En exploitant cette méthode pour détecter et classifier les noeuds du réseau de chemins, un modèle topologique sous forme d'un graphe est construit; la méthode est validée sur une séquence d'images de synthèse. Enfin, dans la dernière partie de la thèse, nous décrivons des résultats expérimentaux obtenus sur le démonstrateur DALA du groupe Robotique et IA du LAAS-CNRS. Le déplacement du robot est contrôlé et guidé par l'information fournie par le système de vision à travers des primitives de déplacement élémentaires (Suivi-Chemin, Suivi-Objet, Suivi-Bordure,...). Le robot se place au milieu du chemin en construisant une trajectoire à partir du contour de cette région navigable. Étant donné que le modèle sémantique de la scène est produit à basse fréquence (de 0,5 à 1Hz) par le module de vision couleur, nous avons intégré avec celui-ci, un module de suivi temporel des bords du chemin (par Snakes), pour augmenter la fréquence d'envoi des consignes (de 5 à 10 Hz) au module de locomotion. Modules de vision couleur et de suivi temporel doivent être synchronisés de sorte que le suivi puisse être réinitialisé en cas de dérive. Après chaque détection du chemin, une trajectoire sur le sol est planifiée et exécutée; les anciennes consignes qui ne sont pas encore exécutées sont fusionnées et filtrées avec les nouvelles, donnant de la stabilité au système.

While semantic web data is machine-understandable and well suited for advanced filtering, in its raw representation it is not conveniently understandable to humans. Therefore, visualization is needed. A core challenge when visualizing the structured but heterogeneous data turned out to be a flexible mapping to Visual Variables. This work deals with a highly flexible, semi-automatic solution with a maximum support of the visualization process, reducing the mapping possibilities to a useful subset. The basis for this is knowledge, concerning metrics and structure of the data on the one hand and available visualization structures, platforms and common graphical facts on the other hand — provided by a novel basic visualization ontology. A declarative, platform-independent mapping vocabulary and a framework was developed, utilizing current standards from the semantic web and the Model-Driven Architecture (MDA).:ABSTRACT S. x 1. INTRODUCTION S. 1 2. VISUALIZATION OF STRUCTURED DATA IN GENERAL S. 4 2.1. Global and Local Interfaces S. 4 2.2. Steps of the Visualization Process S. 4 2.3. Existing Visual Selection Mechanisms S. 6 2.4. Existing Visualizations of Structured Data S. 12 2.5. Categorizing SemVis S. 25 3. REQUIREMENTS FOR A FLEXIBLE VISUALIZATION S. 27 3.1. Actors S. 27 3.2. Use Cases S. 27 4. FRESNEL, A STANDARD DISPLAY VOCABULARY FOR RDF S. 31 4.1. Fresnel Lenses S. 31 4.2. Fresnel Formats S. 33 4.3. Fresnel Groups S. 33 4.4. Primaries (Starting Points) S. 33 4.5. Selectors and Inference S. 34 4.6. Application and Reusability S. 34 4.7. Implementation S. 35 5. A VISUALIZATION ONTOLOGY S. 37 5.1. Describing and Formalizing the Field of Visualization S. 37 5.2. Overview S. 37 5.3. VisualVariable S. 38 5.4. DiscreteVisualValue S. 39 5.5. VisualElement S. 41 5.6. VisualizationStructure S. 42 5.7. VisualizationPlatform S. 42 5.8. PresentationScenario S. 43 5.9. Facts S. 44 6. A NOVEL MAPPING VOCABULARY FOR SEMANTIC VISUALIZATION S. 45 6.1. Overview S. 45 6.2. Mapping S. 46 6.3. PropertyMapping S. 47 6.4. ImplicitMapping S. 48 6.5. ExplicitMapping S. 53 6.6. MixedMapping S. 54 6.7. ComplexMapping S. 55 6.8. Inference S. 58 6.9. Explicit Display of Relations S. 58 6.10. Limitations s. 59 7. A MODEL-DRIVEN ARCHITECTURE FOR FLEXIBLE VISUALIZATION S. 60 7.1. A Model-Driven Architecture S. 61 7.2. Applications of the MDA Pattern S. 62 7.3. Complete System Overview S. 71 7.4. Additional Knowledge of the System S. 72 7.5. Comparison to the Graphical Modelling Framework — GMF S. 77 8. VISUALIZATION PLATFORMS S. 80 8.1. Extensible 3D (X3D) S. 80 8.2. Scalable Vector Graphics (SVG) S. 81 8.3. XHTML + CSS S. 82 8.4. Text S. 82 9. OUTLOOK AND CONCLUSION S. 84 9.1. Advanced Mapping Vocabulary S. 84 9.2. Reusing Standardized Ontologies S. 84 9.3. Enabling Dynamic, Interaction and Animation S. 84 9.4. Implementation and Evaluation S. 85 9.5. Conclusion S. 85 GLOSSARY S. 86 BIBLIOGRAPHY S. 87 A. S. 90 A.1. Schemata S. 90
Während Semantic-Web-Daten maschinenverstehbar und hervorragend filterbar sind, sind sie — in ihrer Rohform — nicht leicht von Menschen verstehbar. Eine Visualisierung der Daten ist deshalb notwendig. Die Kernherausforderung dabei ist eine flexible Abbildung der strukturierten aber heterogenen Daten auf Visuelle Variablen. Diese Arbeit beschreibt eine hochflexible halbautomatische Lösung bei maximaler Unterstützung des Visualisierungsprozesses, welcher die Abbildungsmöglichkeiten, aus denen der Nutzer zu wählen hat, auf eine sinnvolle Teilmenge reduziert. Die Grundlage dafür sind einerseits Metriken und das Wissen über die Struktur der Daten und andererseits das Wissen über verfügbare Visualisierungsstrukturen, -plattformen und bekannte grafische Fakten, welche durch eine neuentwickelte Visualisierungsontologie bereitgestellt werden. Basierend auf Standards des Semantic Webs und der Model-getriebenen Architektur, wurde desweiteren ein deklaratives, plattformunabhängiges Visualisierungsvokabular und -framework entwickelt.:ABSTRACT S. x 1. INTRODUCTION S. 1 2. VISUALIZATION OF STRUCTURED DATA IN GENERAL S. 4 2.1. Global and Local Interfaces S. 4 2.2. Steps of the Visualization Process S. 4 2.3. Existing Visual Selection Mechanisms S. 6 2.4. Existing Visualizations of Structured Data S. 12 2.5. Categorizing SemVis S. 25 3. REQUIREMENTS FOR A FLEXIBLE VISUALIZATION S. 27 3.1. Actors S. 27 3.2. Use Cases S. 27 4. FRESNEL, A STANDARD DISPLAY VOCABULARY FOR RDF S. 31 4.1. Fresnel Lenses S. 31 4.2. Fresnel Formats S. 33 4.3. Fresnel Groups S. 33 4.4. Primaries (Starting Points) S. 33 4.5. Selectors and Inference S. 34 4.6. Application and Reusability S. 34 4.7. Implementation S. 35 5. A VISUALIZATION ONTOLOGY S. 37 5.1. Describing and Formalizing the Field of Visualization S. 37 5.2. Overview S. 37 5.3. VisualVariable S. 38 5.4. DiscreteVisualValue S. 39 5.5. VisualElement S. 41 5.6. VisualizationStructure S. 42 5.7. VisualizationPlatform S. 42 5.8. PresentationScenario S. 43 5.9. Facts S. 44 6. A NOVEL MAPPING VOCABULARY FOR SEMANTIC VISUALIZATION S. 45 6.1. Overview S. 45 6.2. Mapping S. 46 6.3. PropertyMapping S. 47 6.4. ImplicitMapping S. 48 6.5. ExplicitMapping S. 53 6.6. MixedMapping S. 54 6.7. ComplexMapping S. 55 6.8. Inference S. 58 6.9. Explicit Display of Relations S. 58 6.10. Limitations s. 59 7. A MODEL-DRIVEN ARCHITECTURE FOR FLEXIBLE VISUALIZATION S. 60 7.1. A Model-Driven Architecture S. 61 7.2. Applications of the MDA Pattern S. 62 7.3. Complete System Overview S. 71 7.4. Additional Knowledge of the System S. 72 7.5. Comparison to the Graphical Modelling Framework — GMF S. 77 8. VISUALIZATION PLATFORMS S. 80 8.1. Extensible 3D (X3D) S. 80 8.2. Scalable Vector Graphics (SVG) S. 81 8.3. XHTML + CSS S. 82 8.4. Text S. 82 9. OUTLOOK AND CONCLUSION S. 84 9.1. Advanced Mapping Vocabulary S. 84 9.2. Reusing Standardized Ontologies S. 84 9.3. Enabling Dynamic, Interaction and Animation S. 84 9.4. Implementation and Evaluation S. 85 9.5. Conclusion S. 85 GLOSSARY S. 86 BIBLIOGRAPHY S. 87 A. S. 90 A.1. Schemata S. 90

In der vorliegenden Masterarbeit wird erläutert, wie ein klassischer Schema Matcher erweitert wird, um Komplexe Korrespondenzen (many-to-many-Korrespondenzen) und allgemeine Funktionen zwischen zwei Schemata auszudrücken, sowie deren automatische Entdeckung als Erweiterung der herkömmlichen Entdeckung von (1:1)-Korrespondenzen. Der letzte Punkt widmet sich dabei einem Gebiet der Datenintegration, das bisher kaum untersucht wurde, und es werden Ansätze vorgestellt, die für viele Schema Matcher eine Bereicherung darstellen können. Zu diesem Zweckwerden im ersten Teil der Arbeit Komplexe Korrespondenzen und Funktionen im Bereich des Schema Mappings ausführlich vorgestellt.

Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies
Tree species identification forms an integral part of biodiversity monitoring. Locating at-risk species and predicting their distribution is equally as important as tracing invasive alien plant species distributions. The high prevalence of the latter and their destructive impact on the environment is the focus for this thesis. In areas of the world where technology limitations are restrictive, an approach using low-cost, available RGB drone imagery is proposed to train advanced deep learning models to distinguish individual tree species; three dominant species (Pinus elliotti, Eucalyptus grandis and Syzygium cordatum) providing the bulk of sampling data, of which the first two are highly invasive in the region. This study explored the efficacy of utilizing Mask RCNN, an instance segmentation deep neural network, in identifying multiple classes of trees within the same image. In line with the low-cost approach, Google Colaboratory was utilized which drastically lowers the training time necessary and alleviates the need for high GPU systems. The model was trained on imagery from three study areas which were representative of three distinct landscapes: very dense forest, moderately dense forest with overlapping canopies, and open forest. The results indicate decent performance in open forest landscapes where overlapping tree crowns is infrequent with mean Average Precision of 0.71. On the contrary, in a dense forest landscape with many interlocking tree crowns, a mean Average Precision of 0.43 is highly indicative of the model’s poor performance in such environments. The trained network was also observed to have higher confidence scores of detected objects within the open forest study areas as opposed to dense forest.

Background: In the life sciences, the amount of literature and experimental data grows at a tremendous rate. In order to effectively access and integrate these data, biomedical ontologies – controlled, hierarchical vocabularies – are being developed. Creating and maintaining such ontologies is a difficult, labour-intensive, manual process. Many computational methods which can support ontology construction have been proposed in the past. However, good, validated systems are largely missing. Motivation: The biocuration community plays a central role in the development of ontologies. Any method that can support their efforts has the potential to have a huge impact in the life sciences. Recently, a number of semantic search engines were created that make use of biomedical ontologies for document retrieval. To transfer the technology to other knowledge domains, suitable ontologies need to be created. One area where ontologies may prove particularly useful is the search for alternative methods to animal testing, an area where comprehensive search is of special interest to determine the availability or unavailability of alternative methods. Results: The Dresden Ontology Generator for Directed Acyclic Graphs (DOG4DAG) developed in this thesis is a system which supports the creation and extension of ontologies by semi-automatically generating terms, definitions, and parent-child relations from text in PubMed, the web, and PDF repositories. The system is seamlessly integrated into OBO-Edit and Protégé, two widely used ontology editors in the life sciences. DOG4DAG generates terms by identifying statistically significant noun-phrases in text. For definitions and parent-child relations it employs pattern-based web searches. Each generation step has been systematically evaluated using manually validated benchmarks. The term generation leads to high quality terms also found in manually created ontologies. Definitions can be retrieved for up to 78% of terms, child ancestor relations for up to 54%. No other validated system exists that achieves comparable results. To improve the search for information on alternative methods to animal testing an ontology has been developed that contains 17,151 terms of which 10% were newly created and 90% were re-used from existing resources. This ontology is the core of Go3R, the first semantic search engine in this field. When a user performs a search query with Go3R, the search engine expands this request using the structure and terminology of the ontology. The machine classification employed in Go3R is capable of distinguishing documents related to alternative methods from those which are not with an F-measure of 90% on a manual benchmark. Approximately 200,000 of the 19 million documents listed in PubMed were identified as relevant, either because a specific term was contained or due to the automatic classification. The Go3R search engine is available on-line under www.Go3R.org.