An Ikonos II image carried out at the Soure Coastal Plain (Marajó Island) revealed
numerous coastal features over coastal plateaus and coastal plain. The principal goal
of this study is to assess the coastal features mapping from segmentation and
classification techniques. In this paper we present a semi-automated, object-based
method for Ikonos multiespectral image classification, which emphasizes context
information. The study site is compartmentalized in two geomorphologic units: i)
coastal plateaus, with subunits sand roads, built áreas, deforested área with grasses
and urban vegetation ; and ii) coastal plain, with subunits freshmarshesm old Sandy
ridge, beach, mangrove, mangrove scarp, flooded área, water. The validation of
Ikonos image classification shows that the proposed method can delineate coastal
features with more than 90% producer’s and user’s accuracy. The Kappa index is
94% and the global classification accuracy is 95%. The analysis shows an increase is
classification accuracy due to integrate espectral and spatial attributes of the objects
(coastal features) to be classified.
|
