@InProceedings{NepomucenoVaFrSaSiSaDu:2003:ClDaRa,
author = "Nepomuceno, Alcina Maria and Valeriano, Dalton de Morisson and
Freitas, Corina da Costa and Santa Rosa, Ant{\^o}nio Nuno de
Castro and Silva, Nilton Correia da and Santos, Jo{\~a}o Roberto
dos and Dutra, Luciano Vieira",
affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto
Nacional de Pesquisas Espaciais (INPE)} and {Instituto Nacional de
Pesquisas Espaciais (INPE)} and {Universidade de Bras{\'{\i}}lia
(UnB). Departamento da Ci{\^e}ncia da Computa{\c{c}}{\~a}o.}
and {Universidade de Bras{\'{\i}}lia (UnB). Instituto de
Geoci{\^e}ncias.} and {Instituto Nacional de Pesquisas Espaciais
(INPE)} and {Instituto Nacional de Pesquisas Espaciais (INPE)}",
title = "Classifica{\c{c}}{\~a}o de dados de radar na Banda-P utilizando
rede neural artificial para mapeamento de cobertura da terra na
regi{\~a}o de Santar{\'e}m, Par{\'a}",
booktitle = "Anais...",
year = "2003",
editor = "Epiphanio, Jos{\'e} Carlos Neves and Fonseca, Leila Maria
Garcia",
pages = "2249 - 2256",
organization = "Simp{\'o}sio Brasileiro de Sensoriamento Remoto, 11. (SBSR).",
publisher = "INPE",
address = "S{\~a}o Jos{\'e} dos Campos",
keywords = "radar, classification, Artificial Neural Networks, land cover.",
abstract = "The present work is concerned with the potential application of
artificial neural networks for the classification of polarimetric
radar images operating in the P band. The study area is located
near the Tapaj{\'o}s National Forest, in the northern State of
Par{\'a}, Brazil, and comprises regions of dense rain forest
partially disturbed by deforestation and agricultural land use.
The remote sensing data were obtained during the test mission of
the polarimetric imaging radars (P (415 MHz) and X (10 GHz) bands)
of the German company AeroSensing RadarSystem GmbH, promoted by
the Brazilian Army and the National Institute for Space Research
in September 2000. A P band 2.4 km x 7.4 km- image was selected to
assess the capacity of the neural network {"}Fuzzy-ART{"} for the
land cover classes discrimination. Two time-domain filtering
processes were compared regarding their ability to reduce the
speckle noise, both of them operating with neighborhood boxes of
three and five cells. Filtered and non-filtered HH, HV as well VV
P band images were used as inputs by the network to generate
classified images. A confusion matrix based on ground truth data
was employed for the global and partial classification accuracy
analysis, which considered seven land cover classes: exposed soil
(IF), pasture/tillage (PC), recent forest regeneration (RN),
intermediate forest regeneration (RI), old forest regeneration
(RA), very old forest regeneration (RMA), and primary forest (FP).
The results show that the unsupervised neural network-based
classification method was able to accordingly map the land cover
classes in respect to the field observations samples. The results
point to prospective use of data and classification methodology
for fast and accurate radar images interpretation, complying with
the needs of ongoing monitoring and fiscalization activities in
the dense rain forest in a quick and efficient manner.",
conference-location = "Belo Horizonte",
conference-year = "5-10 abr. 2003",
copyholder = "SID/SCD",
isbn = "85-17-00017-X",
language = "Portuguese",
organisation = "Instituto Nacional de Pesquisas Espaciais",
ibi = "ltid.inpe.br/sbsr/2002/11.22.21.04",
url = "http://urlib.net/ibi/ltid.inpe.br/sbsr/2002/11.22.21.04",
targetfile = "16_446.pdf",
type = "Radar: Processamento e Aplica{\c{c}}{\~o}es / Radar: Processing
and Applications",
urlaccessdate = "18 jun. 2024"
}