Título:	Avaliação de imagens polarimétricas do sensor ALOS-2/PALSAR-2 e de técnicas de mineração de dados na classificação de uso e cobertura de terras do Cerrado
Autor(es):	Camargo, Flávio Fortes
Orientador(es):	Sano, Edson Eyji
Assunto:	Sensoriamento remoto Sensores de radar Mapeamento do solo Solo - uso Cerrados - solos Classificação de imagens
Data de publicação:	10-Out-2019
Data de defesa:	7-Dez-2018
Referência:	CAMARGO, Flávio Fortes. Avaliação de imagens polarimétricas do sensor ALOS-2/PALSAR-2 e de técnicas de mineração de dados na classificação de uso e cobertura de terras do Cerrado. 2018. 195 f., il. Tese (Doutorado em Geociências Aplicadas)—Universidade de Brasília, Brasília, 2018.
Abstract:	The Brazilian tropical savanna (Cerrado) occupies an area of approximately 2 million km² (23% of the Brazilian territory) mainly in the Brazilian Central Plateau. The Cerrado vegetation includes forestlands, shrublands and fields that have undergone severe changes with the introduction of extensive and mechanized agricultural production of grains for exportation. Remote sensing technologies have been used to monitor the Cerrado´s vegetation cover using mainly optical satellite images (e.g., the TerraClass and MapBiomas projects). Despite the methodological and technological advances, there are lots of effort to be done using synthetic aperture radar (SAR) satellites to map land use and land cover and to discriminate Cerrados´s phytophysiognomies. In this thesis, two experiments were carried out with ALOS-2/PALSAR-2 SAR images, in two different study areas over the Brazilian Cerrado. The first experiment (Paper # 1) was carried out in a study area (356 km²) located in the northern portion of Brasília, Federal District. The methodological approach proposed in this first experiment combined multiresolution segmentation, object attributes and machine learning procedures. A set of 397 attributes was generated based on the amplitude, HH- and HV-polarized images. These attributes were processed in the WEKA 3.8 software using the J48 (decision tree - DT J48), Random Forest (RF) and Multilayer Perceptron Artificial Neural Network (MLP) classifiers. Classification results attained Kappa indices higher than 0.70, especially the MLP algorithm, with a Kappa index of 0.87. The RF algorithm presented lower performance in comparison with the results presented in the literature, probably due to the reduced number and poor spatial distribution of the training samples. In the second experiment (Paper # 2), a more comprehensive methodology was proposed for the classification of ALOS-2/PALSAR-2 polarimetric SAR images, aiming at the land use and land cover mapping. The study area (3,660 km²) encompassed the mid-east of the Goiás State and the north and northeast of the Federal District. The methodological approach of this second experiment combined polarimetric attributes, multiresolution segmentation, segment attributes and machine learning procedures. A polarimetric pixel-to-pixel classifier (Polarimetric Wishart classifier - PW) was also employed for comparison purposes. The PW classifier is based on distance measures calculated using the Wishart distribution. A set of 125 attributes were generated using multipolarimetric images, including the target decomposition components (van Zyl, Freeman-Durden, Yamaguchi, and Cloude-Pottier procedures), incoherent polarimetric parameters (biomass indices and polarization ratios) and polarized amplitude images (HH, HV, VH, and VV polarizations). These attributes were processed in the WEKA 3.8 software using the Naive Bayes (NB), AD J48, RF, MLP, and Support Vector Machine (SVM) classification algorithms. The RF, MLP, and SVM classifiers presented the best performances and they were considered statistically equal in both proposed scenarios (nine and five thematic classes). Classifiers NB and AD J48 also presented statistically equal results in both scenarios, with AD J48 being more adequate to identify urban areas and natural vegetation coverages. The PW classifier presented the lowest performance among all classifiers. Despite its low performance, the PW classifier presented high potential for classifying forestlands by means of L-band images. The two workflows proposed in this thesis are agile and have potential to be replicated for other satellite images operating at wavelengths other than that from the ALOS-2/PALSAR-2 satellite.
Unidade Acadêmica:	Instituto de Geociências (IG)
Informações adicionais:	Tese (doutorado)—Universidade de Brasília, Instituto de Geociências, Pós-Graduação em Geociências Aplicadas, 2018.
Programa de pós-graduação:	Programa de Pós-Graduação em Geociências Aplicadas e Geodinâmica
Licença:	A concessão da licença deste item refere-se ao termo de autorização impresso assinado pelo autor com as seguintes condições: Na qualidade de titular dos direitos de autor da publicação, autorizo a Universidade de Brasília e o IBICT a disponibilizar por meio dos sites www.bce.unb.br, www.ibict.br, http://hercules.vtls.com/cgi-bin/ndltd/chameleon?lng=pt&skin=ndltd sem ressarcimento dos direitos autorais, de acordo com a Lei nº 9610/98, o texto integral da obra disponibilizada, conforme permissões assinaladas, para fins de leitura, impressão e/ou download, a título de divulgação da produção científica brasileira, a partir desta data.
Aparece nas coleções:	Teses, dissertações e produtos pós-doutorado

Os itens no repositório estão protegidos por copyright, com todos os direitos reservados, salvo quando é indicado o contrário.