Uma abordagem semissupervisionada para classificação de pastagens usando séries temporais de NDVI

• Ano de defesa: 2020
• Autor(a) principal: Oliveira, Evando Natal Fernandes de
• Orientador(a): Laureano, Gustavo Teodoro
• Banca de defesa: Soares, Anderson da Silva, Parente, Leandro Leal, Laureano, Gustavo Teodoro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Ciência da Computação (INF)
• Departamento: Instituto de Informática - INF (RG)
• País: Brasil
• Palavras-chave em Português: Séries temporais; NDVI; Pastagens; Clusterização; Classificação supervisionada; MLMT; MLME; MODIS
• Palavras-chave em Inglês: Time series; Pasture; Clustering; Semisupervised classification; LTMM; LSMM
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/10380

The knowledge and the management of land cover and land use are fundamental in a scenario of food production increasing, due to growth of the world population and change in its eating habits, especially the increase in animal protein consumption, what demands an increase in the herd of cattle and, consequently, in the pasture areas. Remote sensing has been an ally of public managers and the community for a long time, with its abundant data production about the terrestrial surface in different spatial, spectral and temporal resolutions, with particular emphasis on vegetation indices. One of these indices, the NDVI, is calculated and made available from the data generated by the MODIS satellite sensor as a time series, which is one of the most used sources of information for the classification of the most varied types of vegetation. n this work, we present a methodology for classifying pastures that comprises, in a first step, the use of the Linear Temporal Mixture Model -- LTMM, with the final members being obtained from an unsupervised classification method. Secondly, the data are labeled from a pasture map produced by the Processing of Images and Geoprocessing Laboratory of the Federal University of Goi\'as (LAPIG - UFG), which has better spatial resolution than the data generated by MODIS. Then, a classification model is constructed to be applied to the classifying data and its quality is measured by comparison with another pasture map, also produced by LAPIG, with the same spatial resolution than the classified data. The methodology used here presented results with quality compatible with other studies that had purely supervised training approaches for the classification of pastures, using the same data base.