Publicação

Slum mapping : a comparison of single class learning and expert system object-oriented classification for mapping slum settlements in Addis Ababa city, Ethiopia

Ver documento

Detalhes bibliográficos
Resumo:Updated spatial information on the dynamics of slums can be helpful to measure and evaluate the progress of urban upgrading projects and policies. Earlier studies have shown that remote sensing techniques, with the help of very-high resolution imagery, can play a significant role in detecting slums, and providing timely spatial information. The main objective of this thesis is to develop a reliable object-oriented slum identification technique that enables the provision of timely spatial information about slum settlements in Addis Ababa city. It compares the one-class support vector machines algorithm with the expert defined classification rule set in the discrimination of slums, using GeoEye-1 imagery. Two different approaches, called manual and automatic fine-tuning, were deployed to determine the best value of parameters in one-class support vector machines algorithm. The manual fine-tuning of the parameters is done using extensive manual trial. The automatic tuning is done using cross-validation grid search with the overall accuracy as the performance metric. Two regions of study were defined with different landscape compositions, providing different classification scenarios to compare the classification approaches. After image segmentation, twenty predictive variables were computed to characterize the objects in both study areas. An image analyst collected one hundred sample objects of a slum to be used as training for the single-class learner. In parallel, an image analyst has defined a hierarchical rule set to discriminate the class of interest. Results in both study areas indicate that the one-class support vector machine with manual tuning yields higher overall accuracy (97.7% in subset 1, and 92% in subset 2) and requiring much less application effort and computing time than the expert system.
Autores principais:Tesfay, Mulu Weldegebreal
Assunto:Remote sensing Image segmentation Image classification Object-oriented image analysis Single class learning One class support vector machine Expert system classification
Ano:2018
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL

Registos relacionados

Registos relacionados