%0 Conference Proceedings
%4 sid.inpe.br/marte2/2015/06.15.14.56.25
%2 sid.inpe.br/marte2/2015/06.15.14.56.26
%@isbn 978-85-17-0076-8
%F 309
%T Development of dissimilarity functions using stochastic distances for region-based land cover classification: a case study near Tapajós Flona, Pará state, Brazil
%D 2015
%A Dutra, Luciano Vieira,
%A Negri, Rogério Galante,
%A Sant'Anna, Sidnei João Siqueira,
%A Lu, Dengsheng,
%@affiliation Instituto Nacional de Pesquisas Espaciais (INPE)
%@electronicmailaddress dutra@dpi.inpe.br
%E Gherardi, Douglas Francisco Marcolino,
%E Aragão, Luiz Eduardo Oliveira e Cruz de,
%B Simpósio Brasileiro de Sensoriamento Remoto, 17 (SBSR)
%C João Pessoa
%8 25-29 abr. 2015
%I Instituto Nacional de Pesquisas Espaciais (INPE)
%J São José dos Campos
%P 1655-1662
%S Anais
%1 Instituto Nacional de Pesquisas Espaciais (INPE)
%X One recent alternative to standard pixel based classification of remote sensing data, is the region based classification (RBC), which has been proved particularly useful when analyzing high resolution imagery of complex environments, like urban areas. First the imagery is decomposed into homogenous regions, following some criteria, and then each region is classified to one of the classes of interest. Normally, classification is performed by using stochastic distances, which measures the distance of the pixels distribution inside an unknown region and the representative distributions of each class. The class, whose distance is minimum to the unknown region distribution, is assigned to the region, which is known as stochastic minimum distance classification (SMDC). A problem appears when one, or more, class distribution is multi-modal, which violates the Gaussian hypotheses used for classes distributions, degrading the mapping accuracy. This investigation reports the usage of different compositions of the original stochastic minimum distance classifier with the objective of getting less sensitive results for classification, when potentially multi-modal classes are used. The newly developed classifier, called stochastic nearest distance classifier (SNDC), produced the best result when compared with the original classifier and other possible compositions, in a study case near the Tapajós Flona, in Pará state, Brazil. This study also brings, as methodological contribution, a criterion to improve the segmentation phase of RBC methods.
%9 Classificação e mineração de dados
%@language en
%3 p0309.pdf