Michelle Pontes Seixas

Scientific Iniciation Student of

Departamento de Estatística - UFPB - João Pessoa

Advisor: Prof. Ronei Marcos de Moraes.



Project:

Studies on Remote Sensing Image Classification



Abstract


Resumo



Original images:

      

      


RGB composition image of an area of the Tapajós National Forest:

Click on the image for enlarge


Methodology:

For each classifier, we used without rejection option and with rejection option, according each methodology.


Classification Results

    We used an already existing visual classification with eight classes for the area was as reference map to allow a comparison with the results obtained by all classifiers. At Table, we have the best classification for each method and the percentual of correct classification. For a better comparison, the colors are the same in all classifications.

    Click on the image for enlarge

    Classifier Best classification Percentual of correct classification
    Visual reference map
    Paralelepiped 27.68%, with 4.0 standard deviation.
    Maximum Likelihood 58.85%, no rejection option
    Mahalanobis 53.93%, no rejection option
    Minimum Distance 54.18%, no rejection option
    K-NN 32,47%, with K = 10
    Moraes (1998) 86.09%, no option



Departamento de Estatística
Centro de Ciências Exatas e da Natureza
Universidade Federal da Paraíba
Cidade Universitária s/n João Pessoa, PB, Brasil
Telefone: (+55-83) 216-7075
Fax: (+55-83) 216-7117
E-mail address: ronei@de.ufpb.br