@MastersThesis{Marujo:2016:InAtEs,
author = "Marujo, Rennan de Freitas Bezerra",
title = "Influ{\^e}ncia dos atributos espectrais, texturais e fator de
ilumina{\c{c}}{\~a}o na classifica{\c{c}}{\~a}o baseada em
objetos de {\'a}reas cafeeiras",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2016",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2016-02-19",
keywords = "caf{\'e}, classifica{\c{c}}{\~a}o baseada em objetos,
minera{\c{c}}{\~a}o de dados, coffee, object based
classification, data mining.",
abstract = "O caf{\'e}, por ser um importante produto nas
exporta{\c{c}}{\~o}es brasileiras, necessita de constante
monitoramento e pesquisas, para que os sistemas de previs{\~a}o
de safras existentes sejam confi{\'a}veis. Nesta pesquisa foi
avaliado o desempenho da classifica{\c{c}}{\~a}o baseada em
objetos, associada a t{\'e}cnicas de minera{\c{c}}{\~a}o de
dados, aplicada em imagens OLI/\emph{Landsat-8}, com finalidade
de mapeamento de lavouras cafeeiras na microrregi{\~a}o de
Alfenas (MG). Foram feitas tr{\^e}s an{\'a}lises, a primeira
utilizando apenas atributos espectrais, a segunda incluindo
atributos texturais e a terceira, considerando tamb{\'e}m classes
de ilumina{\c{c}}{\~a}o do relevo, extra{\'{\i}}das por meio
do fator de ilumina{\c{c}}{\~a}o. Foram utilizadas seis imagens
multiespectrais OLI/\emph{Landsat-8}, de datas distintas,
referentes a tr{\^e}s diferentes est{\'a}dios fenol{\'o}gicos
da cultura: frutifica{\c{c}}{\~a}o, grana{\c{c}}{\~a}o e
repouso. Al{\'e}m das imagens multiespectrais, foram tamb{\'e}m
utilizados dados da miss{\~a}o SRTM, para determinar as
vari{\'a}veis topogr{\'a}ficas como declividade,
orienta{\c{c}}{\~a}o e o fator de ilumina{\c{c}}{\~a}o do
terreno. Ap{\'o}s corre{\c{c}}{\~a}o atmosf{\'e}rica das
imagens utilizando o m{\'e}todo \emph{Flaash}, aplicou-se o
algoritmo de segmenta{\c{c}}{\~a}o multirresolu{\c{c}}{\~a}o
parametrizado em fator de escala 30, forma 0,6 e compacidade 0,5.
Posteriormente fez-se um processo de minera{\c{c}}{\~a}o de
dados por meio do algoritmo C4.5, o qual gerou {\'a}rvores de
decis{\~a}o para classificar as imagens. A valida{\c{c}}{\~a}o
das classifica{\c{c}}{\~o}es foi feita por meio do M{\'e}todo
de Monte Carlo utilizando como refer{\^e}ncia mapas obtidos por
interpreta{\c{c}}{\~a}o visual. Nas classifica{\c{c}}{\~o}es
feitas utilizando somente atributos espectrais, obteve-se
exatid{\~a}o m{\'e}dia para a classe caf{\'e} de 53\%. Quando
repetiu-se as classifica{\c{c}}{\~o}es, inserindo tamb{\'e}m
atributos texturais e classes de ilumina{\c{c}}{\~a}o do
terreno, a exatid{\~a}o da classe caf{\'e} foi incrementada para
67\%. Em escala municipal a metodologia apresentou melhores
resultados, concedendo exatid{\~a}o para a classe caf{\'e} de
73,83\% no munic{\'{\i}}pio de Machado, que apresenta relevo
acidentado e 82,83\% no munic{\'{\i}}pio de Alfenas, que
trata-se de uma {\'a}rea mais plana. N{\~a}o houve est{\'a}dio
fenol{\'o}gico que proporcionasse maior exatid{\~a}o {\`a}
classe caf{\'e} na classifica{\c{c}}{\~a}o autom{\'a}tica das
imagens OLI/\emph{Landsat-8}. ABSTRACT: Coffee, for being an
important product in Brazilian exportations, needs constant
monitoring and research, so that crop monitoring systems can be
sound and reliable. This research evaluated the performance of an
object based classification associated with data mining techniques
applied in OLI/Landsat-8 images, with the purpose of mapping of
coffee crops in the region of Alfenas, state of Minas Gerais in
Brazil. Three analyzes were made, the first one using only the
spectral attributes; the second including textural attributes and
the third considering also the shaded relief classes. Six
multiespectral images from OLI/Landsat-8 were used, each one of a
different date, relating to three different phenology stages:
frutification, grain formation and rest. In addition to
multispectral images, SRTM data were also used to determine the
topographic variables such as slope, aspect and shaded relief.
After atmospheric correction, the multiresolution segmentation
algorithm were applied, and later its segments became entry to a
data mining process by C4.5 algorithm, which generated decision
trees to classify the images. The accuracy of the classifications
was assessed by the Monte Carlo method using as reference the
images obtained by visual interpretation. In the classification
made using only spectral attributes was obtained an accuracy of
53\% for coffee class. When was inserted textural attributes in
the classification, the accuracy of the coffee class was increased
to 67\%. At the municipal level the methodology presented better
results, providing accuracy of 73.83\% to coffee class in the
municipality of Machado and 82.83\% in Alfenas. There were no
preferential phenology stage that provided greater accuracy to the
coffee class in the automatic classification of OLI/Landsat-8
images.",
committee = "Moreira, Maur{\'{\i}}cio Alves (presidente) and Volpato,
Margarete Marin Lordelo (orientadora) and Formaggio, Antonio
Roberto and Alves, Helena Maria Ramos",
copyholder = "SID/SCD",
englishtitle = "Influence of shaded relief, spectral and textural attributes in
automatic object based classification of coffee areas",
language = "pt",
pages = "96",
ibi = "8JMKD3MGP3W34P/3L3KGLP",
url = "http://urlib.net/ibi/8JMKD3MGP3W34P/3L3KGLP",
targetfile = "publicacao.pdf",
urlaccessdate = "27 abr. 2024"
}