Thesis (MSc (Geography and Environmental Studies))--University of Stellenbosch, 2011. / ENGLISH ABSTRACT: The automation of information extraction from earth observation imagery has become a field of active research. This is mainly due to the high volumes of remotely sensed data that remain unused and the possible benefits that the extracted information can provide to a wide range of interest groups. In this work an earth observation image processing system is presented and profiled that attempts to streamline the information extraction process, without degradation of the quality of the extracted information, for geographic object anomaly detection. The proposed system, implemented as a software application, combines recent research in automating image segment generation and automatically finding statistical classifier parameters and attribute subsets using evolutionary inspired search algorithms. Exploratory research was conducted on the use of an edge metric as a fitness function to an evolutionary search heuristic to automate the generation of image segments for a region merging segmentation algorithm having six control parameters. The edge metric for such an application is compared with an area based metric. The use of attribute subset selection in conjunction with a free parameter tuner for a one class support vector machine (SVM) classifier, operating on high dimensional object based data, was also investigated. For common earth observation anomaly detection problems using typical segment attributes, such a combined free parameter tuning and attribute subset selection system provided superior statistically significant results compared to a free parameter tuning only process. In some extreme cases, due to the stochastic nature of the search algorithm employed, the free parameter only strategy provided slightly better results. The developed system was used in a case study to map a single class of interest on a 22.5 x 22.5km subset of a SPOT 5 image and is compared with a multiclass classification strategy. The developed system generated slightly better classification accuracies than the multiclass classifier and only required samples from the class of interest. / AFIKAANSE OPSOMMING: Die outomatisering van die verkryging van inligting vanaf aardwaarnemingsbeelde het in sy eie reg 'n navorsingsveld geword as gevolg van die groot volumes data wat nie benut word nie, asook na aanleiding van die moontlike bydrae wat inligting wat verkry word van hierdie beelde aan verskeie belangegroepe kan bied. In hierdie tesis word 'n aardwaarneming beeldverwerkingsstelsel bekend gestel en geëvalueer. Hierdie stelsel beoog om die verkryging van inligting van aardwaarnemingsbeelde te vergemaklik deur verbruikersinteraksie te minimaliseer, sonder om die kwaliteit van die resultate te beïnvloed. Die stelsel is ontwerp vir geografiese voorwerp anomalie opsporing en is as 'n sagteware program geïmplementeer. Die program kombineer onlangse navorsing in die gebruik van evolusionêre soek-algoritmes om outomaties goeie beeldsegmente te verkry en parameters te vind, sowel as om kenmerke vir 'n statistiese klassifikasie van beeld segmente te selekteer. Verkennende navorsing is gedoen op die benutting van 'n rand metriek as 'n passings funksie in 'n evolusionêre soek heuristiek om outomaties goeie parameters te vind vir 'n streeks kombinering beeld segmentasie algoritme met ses beheer parameters. Hierdie rand metriek word vergelyk met 'n area metriek vir so 'n toepassing. Die nut van atribuut substel seleksie in samewerking met 'n vrye parameter steller vir 'n een klas steun vektor masjien (SVM) klassifiseerder is ondersoek op hoë dimensionele objek georiënteerde data. Vir algemene aardwaarneming anomalie opsporings probleme met 'n tipiese segment kenmerk versameling, het so 'n stelsel beduidend beter resultate as 'n eksklusiewe vrye parameter stel stelsel gelewer in sommige uiterste gevalle. As gevolg van die stogastiese aard van die soek algoritme het die eksklusiewe vrye parameter stel strategie effens beter resultate gelewer. Die stelsel is getoets in 'n gevallestudie waar 'n enkele klas op 'n 22.5 x 22.5km substel van 'n SPOT 5 beeld geïdentifiseer word. Die voorgestelde stelsel, wat slegs monsters van die gekose klas gebruik het, het beter klassifikasie akkuraathede genereer as die multi klas klassifiseerder.

abstract: Data imbalance and data noise often coexist in real world datasets. Data imbalance affects the learning classifier by degrading the recognition power of the classifier on the minority class, while data noise affects the learning classifier by providing inaccurate information and thus misleads the classifier. Because of these differences, data imbalance and data noise have been treated separately in the data mining field. Yet, such approach ignores the mutual effects and as a result may lead to new problems. A desirable solution is to tackle these two issues jointly. Noting the complementary nature of generative and discriminative models, this research proposes a unified model fusion based framework to handle the imbalanced classification with noisy dataset. The phase I study focuses on the imbalanced classification problem. A generative classifier, Gaussian Mixture Model (GMM) is studied which can learn the distribution of the imbalance data to improve the discrimination power on imbalanced classes. By fusing this knowledge into cost SVM (cSVM), a CSG method is proposed. Experimental results show the effectiveness of CSG in dealing with imbalanced classification problems. The phase II study expands the research scope to include the noisy dataset into the imbalanced classification problem. A model fusion based framework, K Nearest Gaussian (KNG) is proposed. KNG employs a generative modeling method, GMM, to model the training data as Gaussian mixtures and form adjustable confidence regions which are less sensitive to data imbalance and noise. Motivated by the K-nearest neighbor algorithm, the neighboring Gaussians are used to classify the testing instances. Experimental results show KNG method greatly outperforms traditional classification methods in dealing with imbalanced classification problems with noisy dataset. The phase III study addresses the issues of feature selection and parameter tuning of KNG algorithm. To further improve the performance of KNG algorithm, a Particle Swarm Optimization based method (PSO-KNG) is proposed. PSO-KNG formulates model parameters and data features into the same particle vector and thus can search the best feature and parameter combination jointly. The experimental results show that PSO can greatly improve the performance of KNG with better accuracy and much lower computational cost. / Dissertation/Thesis / Doctoral Dissertation Industrial Engineering 2014

abstract: Bank institutions employ several marketing strategies to maximize new customer acquisition as well as current customer retention. Telemarketing is one such approach taken where individual customers are contacted by bank representatives with offers. These telemarketing strategies can be improved in combination with data mining techniques that allow predictability of customer information and interests. In this thesis, bank telemarketing data from a Portuguese banking institution were analyzed to determine predictability of several client demographic and financial attributes and find most contributing factors in each. Data were preprocessed to ensure quality, and then data mining models were generated for the attributes with logistic regression, support vector machine (SVM) and random forest using Orange as the data mining tool. Results were analyzed using precision, recall and F1 score. / Dissertation/Thesis / Masters Thesis Computer Science 2016

Classification of scientific bibliographic data is an important and increasingly more time-consuming task in a “publish or perish” paradigm where the number of scientific publications is steadily growing. Apart from being a resource-intensive endeavor, manual classification has also been shown to be often performed with a quite high degree of inconsistency. Since many bibliographic databases contain a large number of already classified records supervised machine learning for automated classification might be a solution for handling the increasing volumes of published scientific articles. In this study automated classification of bibliographic data, based on two different machine learning methods; Naive Bayes and Support Vector Machine (SVM), were evaluated. The data used in the study were collected from the Swedish research database SwePub and the features used for training the classifiers were based on abstracts and titles in the bibliographic records. The accuracy achieved ranged between a lowest score of 0.54 and a highest score of 0.84. The classifiers based on Support Vector Machine did consistently receive higher scores than the classifiers based on Naive Bayes. Classification performed at the second level in the hierarchical classification system used clearly resulted in lower scores than classification performed at the first level. Using abstracts as the basis for feature extraction yielded overall better results than using titles, the differences were however very small.

Submitted by Fernando Souza (fernando@biblioteca.ufpb.br) on 2017-10-02T13:42:54Z No. of bitstreams: 1 arquivototal.pdf: 2999468 bytes, checksum: 956cae684176d28e417f331d31b46a00 (MD5) / Made available in DSpace on 2017-10-02T13:42:54Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 2999468 bytes, checksum: 956cae684176d28e417f331d31b46a00 (MD5) Previous issue date: 2017-07-29 / Conselho Nacional de Pesquisa e Desenvolvimento Científico e Tecnológico - CNPq / Incisions are a common task in most surgical procedures. Their learning is traditionally done in universities or teaching centers with the use of synthetic materials, animal parts, or, in more advanced stages, in real patients under the supervision of professionals. The use of simulators can contribute in this context of training, since with them it is possible to realistically simulate the materials used, carry out the practice repeatedly and immediately and automatically assess students' performance. Simulators capable of providing evaluation for the incision made in a given procedure are not common, and even those that exist do not have a specific assessment system for this task. The present study aimed to propose and develop an assessment system for surgical incisions simulated with computational methods, identifying the basic components of this process and using appropriate decision models for each of them. For this, the concepts and variables related to this procedure were studied, highlighting their most relevant characteristics and looking for ways to better provide assessment for them. The developed system considers two steps for the assessment of the incision, pre-surgical and surgical. The classical logic was the decision model used for most of the variables, with specific rules to deal with the particularities of each one. In order to evaluate the incision trajectory, the Support Vector Machine model was selected after experiments that compared the accuracy of the assessment of different decision models applied to databases containing rectilinear incision paths. For the validation of the system, metrics for the submental incision, component of the mandibular reconstruction procedure, used in the treatment of mandibular symphysis fractures, which has high prevalence in Brazil and in the world, were obtained and applied in an incision simulation in this region of the body. A computational incision simulation, the conceptualization of the evaluation system, a concrete implementation applied to the problem of the submental incision and conceptual maps, which systematize the knowledge used from different points of view, were produced in this work. It was verified that the assessment system responded adequately, with the classical logic rules and the Support Vector Machine providing results in accordance with the metrics used. Thus, it is observed that the assessment system proposed in this work represents an adequate tool for the use in the training of incision techniques. / As incisões são uma tarefa comum na maioria dos procedimentos cirúrgicos. O aprendizado delas é tradicionalmente feito nas universidades ou centros de ensino com o uso de materiais sintéticos, peças de animais, ou, em estágios mais avançados, em pacientes reais com a supervisão de profissionais. O uso de simuladores pode contribuir neste contexto de treinamento, uma vez que com eles é possível simular de forma realista os materiais utilizados, realizar a prática repetidas vezes e avaliar de forma imediata e automática o desempenho dos estudantes. Simuladores capazes de fornecer avaliação para a incisão feita em determinado procedimento não são comuns, e mesmo os existentes não possuem um método de avaliação específico para esta tarefa. O presente trabalho teve por objetivo propor e desenvolver um sistema de avaliação para incisões cirúrgicas simuladas com métodos computacionais, identificando os componentes básicos deste processo e empregando modelos de decisão adequados para cada um deles. Para isso, foram levantados os conceitos e as variáveis relacionadas a este procedimento, destacando suas características mais relevantes e buscando formas de melhor fornecer avaliação para eles. O sistema desenvolvido considera duas etapas para a avaliação da incisão, pré-cirúrgica e cirúrgica. A lógica clássica foi o modelo de decisão utilizado para a maior parte das variáveis, havendo regras específicas para lidar com as particularidades de cada uma. Para a avaliação da trajetória da incisão foi utilizado o modelo Support Vector Machine, selecionado após a realização de experimentos que compararam a precisão da avaliação de diferentes modelos de decisão aplicados a bancos de dados contendo caminhos de incisões retilíneas. Para a validação do sistema, métricas para a incisão submentoniana, componente do procedimento de reconstrução mandibular, utilizada no tratamento de fraturas na sínfise mandibular, o qual tem alta prevalência no Brasil e no mundo, foram obtidas e aplicadas em uma simulação de incisão nesta região do corpo. Foram produzidos então uma simulação de incisão computacional, a conceitualização do sistema de avaliação, uma implementação concreta aplicada ao problema da incisão submentoniana e mapas conceituais, que sistematizam os conhecimentos utilizados a partir de diferentes pontos de vista. Verificou-se que o sistema de avaliação respondeu adequadamente, com as regras da lógica clássica e a Support Vector Machine provendo resultados em conformidade com as métricas utilizadas. Desta forma, observa-se que o sistema de avaliação proposto neste trabalho representa uma ferramenta adequada para o uso no treinamento de técnicas de incisão.

Made available in DSpace on 2014-12-17T14:54:51Z (GMT). No. of bitstreams: 1 AdrianoAB.pdf: 2240966 bytes, checksum: d9e93de6b9ef6f0023ed591b4d760ff9 (MD5) Previous issue date: 2008-12-02 / The automatic speech recognition by machine has been the target of researchers in the past five decades. In this period have been numerous advances, such as in the field of recognition of isolated words (commands), which has very high rates of recognition, currently. However, we are still far from developing a system that could have a performance similar to the human being (automatic continuous speech recognition). One of the great challenges of searches for continuous speech recognition is the large amount of pattern. The modern languages such as English, French, Spanish and Portuguese have approximately 500,000 words or patterns to be identified. The purpose of this study is to use smaller units than the word such as phonemes, syllables and difones units as the basis for the speech recognition, aiming to recognize any words without necessarily using them. The main goal is to reduce the restriction imposed by the excessive amount of patterns. In order to validate this proposal, the system was tested in the isolated word recognition in dependent-case. The phonemes characteristics of the Brazil s Portuguese language were used to developed the hierarchy decision system. These decisions are made through the use of neural networks SVM (Support Vector Machines). The main speech features used were obtained from the Wavelet Packet Transform. The descriptors MFCC (Mel-Frequency Cepstral Coefficient) are also used in this work. It was concluded that the method proposed in this work, showed good results in the steps of recognition of vowels, consonants (syllables) and words when compared with other existing methods in literature / O reconhecimento autom?tico da voz por m?quinas inteligentes tem sido a meta de muitos pesquisadores nas ?ltimas cinco d?cadas. Neste per?odo, in?meros avan?os foram alcan?ados, como por exemplo no campo de reconhecimento de palavras isoladas (comandos), o qual atualmente apresenta taxas de reconhecimento muito altas. No entanto, ainda se est? longe de desenvolver um sistema que possa ter um desempenho parecido com o ser humano, ou seja, reconhecimento autom?tico de voz em modo cont?nuo. Um dos grandes desafios das pesquisas de reconhecimento de voz cont?nuo ? a grande quantidade de padr?es existentes, pois as linguagens modernas tais como: Ingl?s, Franc?s, Espanhol e Portugu?s possuem aproximadamente 500.000 palavras ou padr?es a serem identificados. A proposta deste trabalho ? utilizar unidades menores do que a palavra tais como: fonemas, difones e s?labas como unidades base para o reconhecimento da voz, visando o reconhecimento quaisquer palavras sem necessariamente utiliz?-las. O objetivo principal deste trabalho ? reduzir a restri??o imposta pela quantidade excessiva de padr?es existentes, ou seja, a quantidade excessiva de palavras. Com o objetivo de validar esta proposta, o sistema foi desenvolvido e testado para o reconhecimento de palavras isoladas no modo dependente do locutor. O sistema apresentado neste trabalho foi desenvolvido com uma l?gica de reconhecimento hier?rquica baseada nas caracter?sticas de produ??o dos fonemas da l?ngua Portuguesa do Brasil. Estas decis?es s?o feitas atrav?s da utiliza??o de redes neurais do tipo M?quinas de Vetor de Suporte agrupadas na forma de M?quinas de C?mite. Os principais descritores do sinal de voz utilizados, foram obtidos atrav?s da Transformada Wavelet Packet. Os descritores MFCC (Mel-Frequency Cepstral Coefficient) tamb?m s?o utilizados neste trabalho. Pode-se concluir que o m?todo proposto apresentou bons resultados nas etapas de reconhecimento de vogais, consoantes (s?labas) e palavras se comparado com outros m?todos existentes na literatura

abstract: With the increase in computing power and availability of data, there has never been a greater need to understand data and make decisions from it. Traditional statistical techniques may not be adequate to handle the size of today's data or the complexities of the information hidden within the data. Thus knowledge discovery by machine learning techniques is necessary if we want to better understand information from data. In this dissertation, we explore the topics of asymmetric loss and asymmetric data in machine learning and propose new algorithms as solutions to some of the problems in these topics. We also studied variable selection of matched data sets and proposed a solution when there is non-linearity in the matched data. The research is divided into three parts. The first part addresses the problem of asymmetric loss. A proposed asymmetric support vector machine (aSVM) is used to predict specific classes with high accuracy. aSVM was shown to produce higher precision than a regular SVM. The second part addresses asymmetric data sets where variables are only predictive for a subset of the predictor classes. Asymmetric Random Forest (ARF) was proposed to detect these kinds of variables. The third part explores variable selection for matched data sets. Matched Random Forest (MRF) was proposed to find variables that are able to distinguish case and control without the restrictions that exists in linear models. MRF detects variables that are able to distinguish case and control even in the presence of interaction and qualitative variables. / Dissertation/Thesis / Ph.D. Industrial Engineering 2013

abstract: Animals learn to choose a proper action among alternatives according to the circumstance. Through trial-and-error, animals improve their odds by making correct association between their behavioral choices and external stimuli. While there has been an extensive literature on the theory of learning, it is still unclear how individual neurons and a neural network adapt as learning progresses. In this dissertation, single units in the medial and lateral agranular (AGm and AGl) cortices were recorded as rats learned a directional choice task. The task required the rat to make a left/right side lever press if a light cue appeared on the left/right side of the interface panel. Behavior analysis showed that rat's movement parameters during performance of directional choices became stereotyped very quickly (2-3 days) while learning to solve the directional choice problem took weeks to occur. The entire learning process was further broken down to 3 stages, each having similar number of recording sessions (days). Single unit based firing rate analysis revealed that 1) directional rate modulation was observed in both cortices; 2) the averaged mean rate between left and right trials in the neural ensemble each day did not change significantly among the three learning stages; 3) the rate difference between left and right trials of the ensemble did not change significantly either. Besides, for either left or right trials, the trial-to-trial firing variability of single neurons did not change significantly over the three stages. To explore the spatiotemporal neural pattern of the recorded ensemble, support vector machines (SVMs) were constructed each day to decode the direction of choice in single trials. Improved classification accuracy indicated enhanced discriminability between neural patterns of left and right choices as learning progressed. When using a restricted Boltzmann machine (RBM) model to extract features from neural activity patterns, results further supported the idea that neural firing patterns adapted during the three learning stages to facilitate the neural codes of directional choices. Put together, these findings suggest a spatiotemporal neural coding scheme in a rat AGl and AGm neural ensemble that may be responsible for and contributing to learning the directional choice task. / Dissertation/Thesis / Ph.D. Electrical Engineering 2014

Submitted by Pedro Barros (pedro.silvabarros@ufpe.br) on 2018-06-26T22:26:10Z No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) DISSERTAÇÃO Caio Bezerra Souto Maior.pdf: 3924685 bytes, checksum: 6968386bf75059f45ee80306322d2a56 (MD5) / Made available in DSpace on 2018-06-26T22:26:10Z (GMT). No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) DISSERTAÇÃO Caio Bezerra Souto Maior.pdf: 3924685 bytes, checksum: 6968386bf75059f45ee80306322d2a56 (MD5) Previous issue date: 2017-02-21 / CAPES / The useful life time of equipment is an important variable related to reliability and maintenance. The knowledge about the useful remaining life of operation system by means of a prognostic and health monitoring could lead to competitive advantage to the corporations. There are numbers of models trying to predict the reliability’s variable behavior, such as the remaining useful life, from different types of signal (e.g. vibration signal), however several could not be realistic due to the imposed simplifications. An alternative to those models are the learning methods, used when exist many observations about the variable. A well-known method is Support Vector Machine (SVM), with the advantage that is not necessary previous knowledge about neither the function’s behavior nor the relation between input and output. In order to achieve the best SVM’s parameters, a Particle Swarm Optimization (PSO) algorithm is coupled to enhance the solution. Empirical Mode Decomposition (EMD) and Wavelets rise as two preprocessing methods seeking to improve the input data analysis. In this paper, EMD and wavelets are used coupled with PSO+SVM to predict the rolling bearing Remaining Useful Life (RUL) from a vibration signal and compare with the prediction without any preprocessing technique. As conclusion, EMD models presented accurate predictions and outperformed the other models tested. / O tempo de vida útil de um equipamento é uma importante variável relacionada à confiabilidade e à manutenção, e o conhecimento sobre o tempo útil remanescente de um sistema em operação, por meio de um monitoramento do prognóstico de saúde, pode gerar vantagens competitivas para as corporações. Existem diversos modelos utilizados na tentativa de prever o comportamento de variáveis de confiabilidade, tal como a vida útil remanescente, a partir de diferentes tipos de sinais (e.g. sinal de vibração), porém alguns podem não ser realistas, devido às simplificações impostas. Uma alternativa a esses modelos são os métodos de aprendizado, utilizados quando se dispõe de diversas observações da variável. Um conhecido método de aprendizado supervisionado é o Support Vector Machine (SVM), que gera um mapeamento de funções de entrada-saída a partir de um conjunto de treinamento. Para encontrar os melhores parâmetros do SVM, o algoritmo de Particle Swarm Optimization (PSO) é acoplado para melhorar a solução. Empirical Mode Decomposition (EMD) e Wavelets são usados como métodos pré-processamento que buscam melhorar a qualidade dos dados de entrada para PSO+SVM. Neste trabalho, EMD e Wavelets foram usadas juntamente com PSO+SVM para estimar o tempo de vida útil remanescente de rolamentos a partir de sinais de vibração. Os resultados obtidos com e sem as técnicas de pré-processamento foram comparados. Ao final, é mostrado que modelos baseados em EMD apresentaram boa acurácia e superaram o desempenho dos outros modelos testados.

Master of Arts / Department of Geography / Jida Wang / Monitoring salt pans is important especially for agricultural management in arid or semi-arid regions because salt pans can negatively affect human life, wildlife, and ecology. Some of the harmful impacts of salt pans are accelerated desertification, cropland loss, economic downturn, wildlife loss, and forced migration of humans and animals due to salt storms. Spectral salt pan indices based upon remotely sensed data (using spectral properties of Landsat-8 imagery) suggested in previous studies vary by location. In other words, the spectral configuration of a salt index for a given location may not be readily applicable to another location due to spatial heterogeneity of salt components across the continental surface. Using Landsat-8 OLI imagery and climate data sets, this study aims to develop a mapping framework which can effectively extract salt pans and salt playas under various spectral conditions in different geographic locations. Based on training samples selected in eight major salt pans/playas in North America, Central Asia, Africa, and Australia, the mapping framework was designed to include the following steps: i) a conservative salt index to highlight potential salt-covered regions, ii) a calibrated support vector machine (SVM) to extract high-salinity areas in the mask regions, and iii) a posterior quality assurance/ quality control (QA/QC) with assistance of auxiliary datasets (e.g., surface slope and land covers) to eliminate commission errors and refine the extracted saltpan areas. The developed mapping framework was validated in the arid endorheic regions across the western United States, with a total area of 699 thousand square kilometers. Both qualitative and quantitative assessments of the results show reliability of the developed framework. The overall accuracy of the extracted salt pans prior to QA/QC is 97%. The final product after QA/QC achieves an overall accuracy of 99.95% and a Kappa statistic of 0.99.According to the results of salt pans areas and endorheic basins areas, it can be concluded that two aforementioned variables of this study are positively correlated to each other, and 1.10 percent of the entire case study area is covered by salt pans. The accuracy of the results suggests a potential that the mapping framework, together with the collected training sample and algorithms, may be applicable to identify salt pan and salt playa regions across the Earth’s land surface.