@Article{FranciscoAlme:2012:EvPeSt,
author = "Francisco, Cristiane Nunes and Almeida, Claudia Maria",
affiliation = "Univ Fed Fluminense, Inst Geociencias, Dept Anal Geoambiental,
Campus Praia Vermelha, BR-24210310 Niteroi, RJ, Brazil. and
{Instituto Nacional de Pesquisas Espaciais (INPE)}",
title = "Evaluating the performance of statistical and textural attributes
for an object-based land cover classification /
Avalia{\c{c}}{\~a}o de desempenho de atributos
estat{\'{\i}}sticos e texturais em uma classifica{\c{c}}{\~a}o
de cobertura da terra baseada em objeto",
journal = "Boletim de Ci{\^e}ncias Geod{\'e}sicas",
year = "2012",
volume = "18",
number = "2",
pages = "302 326",
month = "Apr.-Jun.",
keywords = "Semantic Networks, Images Classification, Data Mining, ALOS SAR
IMAGES.",
abstract = "This paper aim at evaluating the performance of two semantic
networks generated by data mining for classifying land cover using
GEographic Object-Based Image Analysis (GEOBIA). The first one
used statistical and texture attributes, and the second network
employed only statistical attributes. The attributes were
extracted from ALOS/AVNIR images pan-sharpened with ALOS/PRISM.
Relief information provided by the TOPODATA geomorphometric
database was also used as input data. The studied area is Nova
Friburgo County, with an extension of 933 km(2), located in the
mountainous region of Rio de Janeiro State. The Kappa index
obtained by the classification based on statistical and texture
attributes was 0.81, while the result for the classification
derived only from statistical attributes achieved 0.84. These
values corroborate the excellent accuracy of both results. The
statistical hypothesis test between the two indices at 95%
confidence interval demonstrated that there is no difference
between the two classification accuracies. RESUMO Este artigo tem
como objetivo avaliar o desempenho de duas redes sem{\^a}nticas
geradas por minera{\c{c}}{\~a}o de dados para a
classifica{\c{c}}{\~a}o de cobertura da terra por meio de
an{\'a}lise de imagens baseada em objetos geogr{\'a}ficos
(GEographic Object-Based Image Analysis GEOBIA). Para isto, uma
rede utilizou-se de descritores estat{\'{\i}}sticos e texturais,
e a outra, apenas de descritores estat{\'{\i}}sticos. A base de
dados foi constitu{\'{\i}}da de imagens ALOS/AVNIR fusionadas
com imagens ALOS/PRISM e dados de relevo provenientes do banco de
dados TOPODATA. A {\'a}rea de estudo corresponde ao
munic{\'{\i}}pio de Nova Friburgo, com 933 kmē, localizado na
regi{\~a}o serrana do estado do Rio de Janeiro. O {\'{\i}}ndice
Kappa alcan{\c{c}}ado pela classifica{\c{c}}{\~a}o baseada em
{\'a}rvore de decis{\~a}o composta por descritores
estat{\'{\i}}sticos e texturais foi de 0,81, enquanto que este
valor para a classifica{\c{c}}{\~a}o derivada apenas de
descritores estat{\'{\i}}sticos foi de 0,84. Considerando os
{\'{\i}}ndices alcan{\c{c}}ados, conclui-se que ambos os
resultados apresentam excelente qualidade quanto {\`a}
acur{\'a}cia da classifica{\c{c}}{\~a}o. O teste de
hip{\'o}tese entre os dois {\'{\i}}ndices mostra, com
n{\'{\i}}vel de signific{\^a}ncia de 5%, que n{\~a}o h{\'a}
diferen{\c{c}}as entre as duas classifica{\c{c}}{\~o}es quanto
{\`a} acur{\'a}cia. Palavras-Chave: Redes Sem{\^a}nticas;
Classifica{\c{c}}{\~a}o de Imagens; Minera{\c{c}}{\~a}o de
Dados; ALOS. ABSTRACT This paper aim at evaluating the performance
of two semantic networks generated by data mining for classifying
land cover using GEographic Object-Based Image Analysis (GEOBIA).
The first one used statistical and texture attributes, and the
second network employed only statistical attributes. The
attributes were extracted from ALOS/AVNIR images pan-sharpened
with ALOS/PRISM. Relief information provided by the TOPODATA
geomorphometric database was also used as input data. The studied
area is Nova Friburgo County, with an extension of 933 kmē,
located in the mountainous region of Rio de Janeiro State. The
Kappa index obtained by the classification based on statistical
and texture attributes was 0.81, while the result for the
classification derived only from statistical attributes achieved
0.84. These values corroborate the excellent accuracy of both
results. The statistical hypothesis test between the two indices
at 95% confidence interval demonstrated that there is no
difference between the two classification accuracies.",
doi = "10.1590/S1982-21702012000200008",
url = "http://dx.doi.org/10.1590/S1982-21702012000200008",
issn = "1413-4853",
language = "en",
targetfile = "08-1.pdf",
urlaccessdate = "17 maio 2024"
}