@InProceedings{MarettoKörCasFonSan:2016:SpAtSe,
author = "Maretto, Raian and K{\"o}rting, Thales Sehn and Castejon,
Emiliano Ferreira and Fonseca, Leila Maria Garcia and Santos,
Rafael Duarte Coelho dos",
affiliation = "{} and {Instituto Nacional de Pesquisas Espaciais (INPE)} and
{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto
Nacional de Pesquisas Espaciais (INPE)} and {Instituto Nacional de
Pesquisas Espaciais (INPE)}",
title = "Spectral attributes selection based on data mining for remote
sensing image classification",
year = "2016",
organization = "Workshop de Computa{\c{c}}{\~a}o Aplicada, 16. (WORCAP)",
abstract = "Remote sensing images are a rich source of information for
studying large-scale geographic areas. The increased accessibility
of the new generation high-spatial resolution multispectral
sensors has improved the level of complexity required in the
analysis techniques. In particular, many traditional per-pixel
analysis may not be suitable to high-spatial resolution imagery,
due to its high-frequency components and the horizontal layover
caused by off-nadir look angles [Im et al. 2008]. Aiming to
overcome this problem, in the last decades, several approaches and
platforms have been developed with algorithms that consider
contextual information and pixel region properties [K{\"o}rting
et al. 2013; Syed et al. 2005; Walter 2004]. Current software can
extract several statistical, spatial, color, texture or
topological attributes. However, most of them often do not help to
distinguish between the classes of interest, due to its high
correlation. Thus, the attributes selection phase often relies on
ad hoc decisions about what of them can better describe the
classes. The huge number of attributes available makes a detailed
exploratory time-consuming and dependent on expertise
[K{\"o}rting et al. 2013]. Many works have proved that data
mining techniques can be useful to this purpose [Dash and Liu
1997; Kohavi and Kohavi 1997; Laliberte et al. 2012]. In this
context, the main objective of this work is to analyze the
correlation of the spectral attributes between a set of classes of
interest, in order to verify what of them best distinguish these
classes. A case study is presented over a small region of the city
of S{\~a}o Jos{\'e} dos Campos, using a WorldView-2 image. It is
important to emphasize that although this study is in a
preliminary stage, the results are promising and reached
improvements in the accuracy of the classification, even as a good
reduction in the computational time.",
conference-location = "S{\~a}o Jos{\'e} dos Campos, SP",
conference-year = "25-26 out.",
language = "en",
targetfile = "maretto_spectral.pdf",
urlaccessdate = "03 maio 2024"
}