| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 8.23 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Existem diversos trabalhos que comparam o método de classificação baseada em pixels com o método de classificação orientada ao objeto que consideram melhores resultados neste último, quando associado aos métodos pixel-a-pixel, sendo esta uma importante ferramenta para a classificação e estudos eficientes do uso do solo (Antunes, 2003).
Os objetos nas imagens são representações reais do terreno e citando Antunes, “Desta maneira, o contexto espacial é descrito em termos de relações topológicas entre os objetos”, ou seja, relações entre as diferentes propriedades de estudo relativamente à sua disposição assim como dos diferentes elementos de um conjunto. Assim considera-se que os objetos homogéneos são representados não apenas pela sua assinatura espectral, mas também pela sua textura, e que a segmentação da imagem em diferentes escalas pode levar à criação de uma rede hierárquica relacionando objetos maior a objetos menores. A Deteção Remota desempenha um papel importante no estudo da biosfera, permitindo realizar medições em diferentes escalas e explorar os dados existentes de modo a melhorar e completar a informação, permitindo uma análise da extensão e localização das áreas urbanas e da distribuição espacial dos diferentes tipos de uso e cobertura do solo.
Nesta dissertação foi utilizada a classificação orientada ao objeto que, ao invés da classificação dos pixels de forma individual quanto às suas características espectrais, interpreta a informação dos objetos originados pelo processo de segmentação da imagem, dividindo esta em segmentos homogéneos, que pode ter como base a sua resposta espectral média, a variância, as dimensões, a forma, e a textura, permitindo a aquisição de informação mais precisa e detalhada (Kux, 2009).
O motivo da escolha deste tipo de classificador deve-se ao fato de os objetos de estudo não resultarem de uma resposta espectral clara, sendo representados mais facilmente pela sua forma ao invés do valor do pixel que se confunde com outras classes. Com esta dissertação pretende-se contribuir para o estudo de identificação de pneus ao ar livre através de métodos automáticos, tendo como base a classificação de imagem orientada ao objeto, utilizando fotografias aéreas de alta resolução, com o intuito de ajudar a preservar o meio ambiente e a saúde pública.
There are several studies that compare the pixel-based classification method with the object-oriented classification method that consider better results in the last one, when associated with pixel-by-pixel methods, which is an important tool for the classification and efficient studies of use soil (Antunes, 2003). The objects in the images are real representations of the field and quoting Antunes, “In this way, the spatial context is described in terms of topological relationships between the objects”, that is, relationships between the different study properties in relation to their disposition as well as the different elements of a set. There for it considered that homogeneous objects represented not only by their spectral signature, but also by their texture, and that the segmentation of the image at different scales can lead to the creation of a hierarchical network relating larger objects to smaller objects. Remote Sensing plays an important role in the study of the biosphere, allowing measurements at different scales and exploring existing data in order to improve and complete the information and allows an analysis of the extent and location of urban areas and the spatial distribution of different types of land use and cover. In this dissertation, object-oriented classification was used, which, instead of classifying pixels individually in terms of their spectral characteristics, interprets the information of objects originated by the image segmentation process, dividing it into homogeneous segments, which can be based on its average spectral response, variance, dimensions, shape, and texture, allowing the acquisition of more accurate and detailed information (Kux, 2009). The reason for choosing this type of classifier is that the objects of study do not result from a clear spectral response, easily represented by their shape instead of the pixel value that is confused with other classes. This dissertation intends to contribute to the study of tire identification in the open-air using automatic methods, based on object-oriented image classification, using high-resolution aerial photographs in order to help preserve the environment and public health.
There are several studies that compare the pixel-based classification method with the object-oriented classification method that consider better results in the last one, when associated with pixel-by-pixel methods, which is an important tool for the classification and efficient studies of use soil (Antunes, 2003). The objects in the images are real representations of the field and quoting Antunes, “In this way, the spatial context is described in terms of topological relationships between the objects”, that is, relationships between the different study properties in relation to their disposition as well as the different elements of a set. There for it considered that homogeneous objects represented not only by their spectral signature, but also by their texture, and that the segmentation of the image at different scales can lead to the creation of a hierarchical network relating larger objects to smaller objects. Remote Sensing plays an important role in the study of the biosphere, allowing measurements at different scales and exploring existing data in order to improve and complete the information and allows an analysis of the extent and location of urban areas and the spatial distribution of different types of land use and cover. In this dissertation, object-oriented classification was used, which, instead of classifying pixels individually in terms of their spectral characteristics, interprets the information of objects originated by the image segmentation process, dividing it into homogeneous segments, which can be based on its average spectral response, variance, dimensions, shape, and texture, allowing the acquisition of more accurate and detailed information (Kux, 2009). The reason for choosing this type of classifier is that the objects of study do not result from a clear spectral response, easily represented by their shape instead of the pixel value that is confused with other classes. This dissertation intends to contribute to the study of tire identification in the open-air using automatic methods, based on object-oriented image classification, using high-resolution aerial photographs in order to help preserve the environment and public health.
Descrição
Palavras-chave
Deteção Remota Cassificação de imagem Segmentação Pneus
