Open Journal Systems

CLASSIFICAÇÃO DE AMBIENTE CILIAR BASEADA EM ORIENTAÇÃO A OBJETO EM IMAGENS DE ALTA RESOLUÇÃO ESPACIAL

Alzir Felippe Buffara Antunes

Abstract



A classificação baseada em contexto é um importante campo de estudos
no processamento digital de imagens. A inserção do conhecimento dentro do
processo de classificação da vegetação é uma das formas atualmente utilizadas pela
comunidade de sensoriamento remoto a fim de incrementar a qualidade da
classificação. O presente estudo é embasado no algoritmo de segmentação multiresolução,
FNEA( fractal net evolution approach), que permite segmentar uma
imagem em diferentes níveis hierárquicos, a inserção do contexto é realizada
através de relações entre os objetos. A área de estudo escolhida foi uma área rural
no município de Nova Esperança-Pr, onde se pretendeu mapear tipologias vegetais
no ambiente ciliar do Rio Paracatu. Com a execução deste trabalho foi possível
apresentar uma proposta de classificação digital de imagem de alta resolução
baseada em orientação a objeto do ambiente ciliar para a área de estudo adotada. Os
objetos oriundos do processo de segmentação multi-resolução permitiram a criação
de diferentes níveis de segmentos, o que pode sustentar uma hierarquia entre
objetos e subobjetos. Esta hierarquia foi o fundamento para a estruturação de uma
rede semântica, baseada no conhecimento. A classificação foi realizada por lógica
fuzzy através de descritores de forma, textura e relações entre objetos e subobjetos.
Foram avaliadas as diferentes opções de classificação dos objetos, tais como rede
hierárquica, o processo seletivo e a mudança de especialista. Analisou-se ainda o
grau de detalhamento das classes e subclasses que os diferentes modelos de classificação apresentam (potencial de discriminação). A acurácia da classificação
foi baseada nas ambigüidades geradas a partir da classificação fuzzy.

Classifying ciliar environment based upon on orientation to objects in high spatial
resolution

Abstract


Context base classification plays an important role in digital image
analysis. The insertion of the knowledge base in the digital classification process is
one of current forms used by the community of remote sensing for vegetation
mapping, in order to develop quality of the classification. Advanced image
segmentation techniques (FNEA: fractal net evolution approach), was used in this
study to provide the context introduction. The tested site was an agricultural area
near the city of Nova Esperança-Pr, wherein was mapped the riparian vegetation
along the Paracatu River. This work tried to present a proposal for high resolution
image classification of the riparian environment based on object oriented analysis.
The objects are derived by means of multiresolution segmentation. It allows a
creation of different levels of segments supporting a hierarchical structure,
generating spatial relations between objects and sub-objects. This hierarchy was the
bedding for the semantic network. The knowledge base was the basis for the
semantics. The classification was based on fuzzy rules by means of descriptors
such as: form, texture and relations between objects and sub-objects. Different
approaches of classification were assessed: semantic network, selective and context
change classification. It also evaluated the degree of detailing of the classes and
subclasses in different levels of segmentation (thematic resolution). The analysis of
classification accuracy relied upon ambiguities generated by fuzzy rules.