Utilização de algoritmos de classificação em uma imagem IKONOS para o mapeamento de feições terrestres: um estudo de caso

• Ano de defesa: 2006
• Autor(a) principal: Mendonça, Luis Eduardo Ribeiro de
• Orientador(a): Soares, Vicente Paulo
• Banca de defesa: Dias, Herly Carlos Teixeira , Vieira, Carlos Antonio Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado em Ciência Florestal
• Departamento: Manejo Florestal; Meio Ambiente e Conservação da Natureza; Silvicultura; Tecnologia e Utilização de
• País: BR
• Palavras-chave em Português: Algoritmos de classificação; Imagem IKONOS; Cobertura terrestre; Mapeamento; Sensoriamento remoto
• Palavras-chave em Inglês: Classification algorithms; IKONOS image; Ground features; Mapping; Remote sensing
• Área do conhecimento CNPq: CNPQ::CIENCIAS AGRARIAS::RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL::MANEJO FLORESTAL
• Link de acesso: http://locus.ufv.br/handle/123456789/3172

The present work consisted in classifying part of an IKONOS II image, from the municipality of Serra do Salitre, Minas Gerais State, Brazil, using six classification algorithms, three of them by pixels and three by regions, to map the ground cover. The algorithms were: Maximum Likelyhood, Minimum Euclydian Distance, Neural Nets, Bhattacharya, Isoseg and Hystogram. The calculations of the accuracy classification, made by the Kappa index, were performed with 250, 500, 750 and 1,000 pixels, to check any accuracy loss if smaller number of samples were used. For this check up it was verified if there were significant statistical differences among the kappa indexes found after the classifications or not. The results indicated that the classifications generated by the algorithms that work with a segmented image generated thematic images with a much higher visual quality than those that did not use this method. Besides, there was not a significant statistical difference, at 95% probability, among the kappa indexes, when the size of the test samples were varied. There was a significant statistical difference, at 95% probability, when comparing the kappa indexes generated by the classifications by pixel and by region. This did not occur when comparing the three algorithms by pixels with one another, and the three algorithms by regions also between each other. The classification algorithms by pixels showed values of kappa indexes considered fairly good while the algorithms by regions showed excellent results, with the Isoseg obtaining the best indexes. The increase in the size of the samples of the tests for the evaluation of the classified images did not statistically show improvements in the kappa accuracy index calculation.