@MastersThesis{Nepomuceno:2003:UsReNe,
author = "Nepomuceno, Alcina Maria",
title = "Uso de rede neural artificial n{\~a}o supervisionada na
classifica{\c{c}}{\~a}o de dados de radar na Banda-P para
mapeamento de cobertura da terra em floresta tropical",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2003",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2003-03-28",
keywords = "Imagem de radar, classifica{\c{c}}{\~a}o de imagens,
reconhecimento de padr{\~a}o, redes neurais, algoritmos
gen{\'e}ticos, cobertura da terra, processamento de imagem,
sensoriamento remoto, imaging radar, p band, image classification,
pattern recognition, neural networks, genetic algorithms, land
cover, image processing, remote sensing.",
abstract = "Apresenta-se uma avalia{\c{c}}{\~a}o sobre as propriedades
discriminat{\'o}rias de dados de radar na Banda-P para o
mapeamento da cobertura da terra usando a rede neural artificial
n{\~a}o supervisionada Fuzzy-ART (Teoria da Resson{\^a}ncia
Adaptativa). A {\'a}rea de estudo situa-se pr{\'o}xima {\`a}
Floresta Nacional do Tapaj{\'o}s, no Estado do Par{\'a}, Brasil.
Os dados de radar foram obtidos durante a miss{\~a}o realizada
pela empresa alem{\~a} AeroSensing RadarSystem GmbH, em setembro
de 2000. Foi selecionada uma faixa de imageamento 2,4 km x 7,4 km
para o estudo. Os par{\^a}metros de entrada para a rede Fuzzy-ART
foram otimizados por algoritmo gen{\'e}tico. Foram investigadas
as efici{\^e}ncias dos filtros Map Gamma (5x5) e a
combina{\c{c}}{\~a}o dos filtros Frost e Mediana (3x3) para
redu{\c{c}}{\~a}o do efeito do ru{\'{\i}}do speckle. As
seguintes imagens foram avaliadas individualmente e combinadas
duas a duas: retroespalhamento nas polariza{\c{c}}{\~o}es HH,
HV, VV, Se{\c{c}}{\~a}o Transversa M{\'e}dia (STM), e os
{\'{\i}}ndices biof{\'{\i}}sicos {\'{\I}}ndice de Biomassa
(BMI), {\'{\I}}ndice de Estrutura do Dossel (CSI) e
{\'{\I}}ndice de Espalhamento Volum{\'e}trico (VSI).
Examinou-se tamb{\'e}m a combina{\c{c}}{\~a}o HH/HV/VV. Os
padr{\~o}es discriminados pela rede neural foram relacionados com
as classes de cobertura da terra de locais previamente observados
em trabalho de campo. As oito classes de refer{\^e}ncia s{\~a}o:
Solo Exposto (SE), Pasto/Cultivo (PC), Regenera{\c{c}}{\~a}o
Nova (RN), Regenera{\c{c}}{\~a}o Intermedi{\'a}ria (RI),
Regenera{\c{c}}{\~a}o Antiga (RA), Regenera{\c{c}}{\~a}o Muito
Antiga (RMA), Floresta Prim{\'a}ria (FP), e V{\'a}rzea (VA). Um
conjunto de amostras de refer{\^e}ncia foi utilizado para
identificar a classe a que os padr{\~o}es pertencem e outro para
calcular a exatid{\~a}o global e o {\'{\i}}ndice Kappa. A
discrimina{\c{c}}{\~a}o de oito classes de cobertura da terra
n{\~a}o foi satisfat{\'o}ria. A melhor exatid{\~a}o global
(56%) foi obtida a partir da lassifica{\c{c}}{\~a}o da imagem
STM. Baseado no grau de confus{\~a}o entre as classes de
refer{\^e}ncia foram realizadas combina{\c{c}}{\~o}es entre
classes e entre seus correspondentes padr{\~o}es para cinco e
quatro classes. Os melhores resultados de exatid{\~a}o global
foram obtidos na discrimina{\c{c}}{\~a}o de quatro classes
(SE/PC; RN/RI; FP/RMA/RA e VA). As seguintes exatid{\~o}es
globais foram obtidas para as imagens classificadas
individualmente: 84%, 73%, 78%, 83%, 74%, 79%, e 76% para HH, HV,
VV, STM, BMI, CSI e VSI, respectivamente. Foram obtidos os
seguintes resultados para as classifica{\c{c}}{\~o}es das
combina{\c{c}}{\~o}es de imagens: 84,9%, 84,5% 83,7%,
81,2%,79,6%, 76,5%, 74,4%, e 72,8% para CSI/HV, HH/HV/VV, HH/HV,
HH/VV, CSI/VV, VSI/HV, BMI/HV e VV/HV, respectivamente. Como
resultado geral das an{\'a}lises, o melhor resultado (84,9%) foi
obtido a partir da combina{\c{c}}{\~a}o das imagens CSI e HV
filtradas com o filtro Map Gamma para a discrimina{\c{c}}{\~a}o
das classes SE/PC, RN/RI,FP/RMA/RA e VA. Conclui-se que a
utiliza{\c{c}}{\~a}o das imagens co-polarizadas e com
polariza{\c{c}}{\~a}o cruzada combinadas contribui para uma
melhora no resultado das classifica{\c{c}}{\~o}es, e que a
aplicabilidade dos dados da Banda-P para avalia{\c{c}}{\~a}o da
cobertura da terra em paisagens de Floresta Tropical {\'e}
somente confi{\'a}vel para classes de cobertura da terra
amplamente definidas. ABSTRACT: The applicability of P-band radar
data for land cover mapping using the unsupervised artificial
neural network Fuzzy-ART (Adaptive Resonance Theory) is evaluated.
The study area is located near Tapaj{\'o}s National Forest in the
State of Para, Brazil. The radar data was acquired during an
airborne mission conducted by AeroSensing RadarSystem GmbH in
september 2000. A 2.4 km x 7.4 km image strip was selected for the
study. The input parameters for the neural network Fuzzy-ART were
optimized by genetic algorithm. It was investigated the speckle
reduction efficiencies of Map Gamma filter (5x5 pixels) and the
combination of Frost and Median filters (3x3 pixels). The
following images were analyzed individually and combined in pairs:
backscatter in the polarizations HH, HV, VV, Average cross section
(ACS), and the biophysical indices Biomass Index (BMI), Canopy
Structure Index (CSI) and Volume Scattering Index (VSI). The
combination HH/HV/VV was also evaluated. The clusters
discriminated by the neural network were related with the land
cover classes of sites previously observed in field work. The
eight reference classes are: Bare Soil (BS), Pasture and
Agriculture (PA), Upland Forest Regrowth - Pioneer Stages (R1),
Upland Forest Regrowth Early Intermediate Stages (R2), Upland
Forest Regrowth Late Intermediate Stages (R3), Upland Forest
Regrowth Advanced Stages (R4), Primary Upland Forest (PF) and
Primary Floodplain Forest (FF). Part of the reference data set was
used for cross tabulation to map unsupervised clusters set onto
the land cover class set and the other part for estimating the
global accuracy and the Kappa coefficient. The discrimination of
the eight land cover classes was not satisfactory. Best global
accuracy (56%) was obtained with PT. Based on the degree of
confusion among reference classes, the combinations of classes and
corresponding clusters were reduced to five and four classes. The
best results of global accuracy were obtained in the
discrimination of four classes (BS/PA; R1/R2; R3/R4/FP and FF).
The following global accuracies were obtained for the individually
classified images: 84%, 73%, 78%, 83%, 74%, 79%, and 76% for HH,
HV, VV, PT, BMI, CSI and VSI, respectively. It was obtained the
following global accuracies for the classifications of combined
images: 84,9%, 84,5% 83,7%, 81,2%, 79,6%, 76,5%, 74,4%, and 72,8%
for CSI/HV, HH/HV/VV, HH/HV, HH/VV, CSI/VV, VSI/HV, BMI/HV and
VV/HV, respectively. As a general result of the analyses, the best
result (global accuracy of 84,9%) was obtained with the
combination of CSI and HV pre-filtered with the Map Gamma filter
for the discrimination of the classes BS/PA; R1/R2; R3/R4/FP and
FF. It was concluded that the utilization of co-polarized and
cross-polarized images contributes for the improvement of the
classification result, and that the applicability of P-band radar
data for land cover assessment in tropical forest landscape is
only reliable for broadly defined land cover classes.",
committee = "Santos, Jo{\~a}o Roberto dos (presidente) and Freitas, Corina da
Costa (orientadora) and Valeriano, Dalton de Morisson (orientador)
and Dutra, Luciano Vieira and Hemerly, Elder Moreira",
copyholder = "SID/SCD",
englishtitle = "P-Band radar data classification by neural network for Amazonin
land cover assessment",
language = "pt",
pages = "197",
ibi = "6qtX3pFwXQZ3P8SECKy/y58ea",
url = "http://urlib.net/ibi/6qtX3pFwXQZ3P8SECKy/y58ea",
targetfile = "paginadeacesso.html",
urlaccessdate = "12 maio 2024"
}