Made available in DSpace on 2016-12-12T20:22:54Z (GMT). No. of bitstreams: 1 Juliana P Detroz.pdf: 11151863 bytes, checksum: f8e2db7d1e13c674adf28e9484a35d9d (MD5) Previous issue date: 2016-02-26 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / With the emerging hydric crisis, water shortage has been a great global concern. Water supply companies have been increasingly looking for solutions to reduce water wastage and many efforts have been made aiming to promote a better management of this resource. Fraud detection is one of these actions, as the irregular violations are usually held precariously, thus, causing leaks. Hidden and apparent leakage is a major cause of the high water loss rates. In this context, the use of technology in order to automate the identification of potential frauds can be an important support tool to avoid water waste. In this sense, this research aims to apply pattern recognition techniques in the implementation of an automated detection of suspected irregularities cases in water meters, through image analysis. We considered as a potential fraud when there is evidences of violations and seals absences. The proposed computer vision system is composed by three steps: the detection of the water meter location, obtained by OPF classifier and HOG descriptor, detecting the seals through morphological image processing and segmentation methods; and the classification of frauds, in which the condition of the water meter seals is assessed. We validated the proposed framework using a dataset containing images of water meter inspections. The water meter detection solution (HOG+OPF) achieved an average accuracy of 89.03%, showing superior results than SVM (linear and RBF). A comparative analysis of 12 feature descriptors (color and texture) was performed on the classification of the seals condition step. The results of these methods were evaluated individually and also combined, reaching average accuracy up to 81.29%. We concluded that the use of a computer vision system is a promising strategy and has potential to benefit and support the analysis of fraud detection. / Em tempos de racionamento dos recursos hídricos, o desperdício de água tem sido um tema de relevância mundial. Os vazamentos ocultos e aparentes são uma das principais causas dos elevados índices de perdas de água tratada. Esforços são despendidos pelas companhias de saneamento a fim de reduzir as perdas, sendo o combate às fraudes uma destas ações. Neste contexto, o uso da tecnologia para automatizar a identificação de fraude mostra-se uma importante ferramenta de apoio no combate ao desperdício. Esta pesquisa tem como objetivo aplicar técnicas de reconhecimento de padrões na detecção automatizada de casos suspeitos de irregularidades em hidrômetros. No escopo deste trabalho foram consideradas suspeitas de fraude as violações e ausências de lacres. A abordagem proposta visa, através de um sistema de visão computacional, auxiliar no combate a fraudes em hidrômetros e, consequentemente, evitar o desperdício de água associado a estas. Para isto, a execução do sistema proposto é dividida em três etapas: detecção do hidrômetro, fazendo uso do classificador OPF e descritor HOG; a detecção da área estimada dos lacres, obtida pela aplicação de métodos de processamento morfológico e segmentação; e a classificação das fraudes a partir da condição dos lacres do hidrômetro. A validação foi executada utilizando-se um conjunto de imagens de fiscalizações. Na primeira etapa, a solução utilizando o classificador OPF alcançou taxa de acerto média de 89, 03%, sendo superior a resultados dos métodos SVM linear e RBF. Para a classificação da condição dos lacres, realizou-se uma análise comparativa de 12 descritores de imagem, de cor e textura, sendo avaliados os resultados individuais e combinados, atingindo taxas de acerto média de até 81, 29%. Com isto, pode-se concluir que o uso de um sistema especialista de visão computacional para o problema de detecção de fraudes é uma estratégia promissora e com potencial para beneficiar a análise e o suporte à tomada de decisões.

Many real-world problems like hand-written digit recognition or semantic scene classification are treated as multiclass or multi-label classification prob-lems. Solutions to these problems using support vector machines (SVMs) are well studied in literature. In this work, we focus on building sparse max-margin classifiers for multiclass and multi-label classification. Sparse representation of the resulting classifier is important both from efficient training and fast inference viewpoints. This is true especially when the training and test set sizes are large.Very few of the existing multiclass and multi-label classification algorithms have given importance to controlling the sparsity of the designed classifiers directly. Further, these algorithms were not found to be scalable. Motivated by this, we propose new formulations for sparse multiclass and multi-label classifier design and also give efficient algorithms to solve them. The formulation for sparse multi-label classification also incorporates the prior knowledge of label correlations. In both the cases, the classification model is designed using a common set of basis vectors across all the classes. These basis vectors are greedily added to an initially empty model, to approximate the target function. The sparsity of the classifier can be controlled by a user defined parameter, dmax which indicates the max-imum number of common basis vectors. The computational complexity of these algorithms for multiclass and multi-label classifier designisO(lk2d2 max), Where l is the number of training set examples and k is the number of classes. The inference time for the proposed multiclass and multi-label classifiers is O(kdmax). Numerical experiments on various real-world benchmark datasets demonstrate that the proposed algorithms result in sparse classifiers that require lesser number of basis vectors than required by state-of-the-art algorithms, to attain the same generalization performance. Very small value of dmax results in significant reduction in inference time. Thus, the proposed algorithms provide useful alternatives to the existing algorithms for sparse multiclass and multi-label classifier design.