Semantic network applied to EKONOS-2 image for the delineation of Araucaria angustifolia (Brazilian pine) crowns in a Mixed Ombrophilous Forest

摘要

This work shows the treatment of IKONOS-2 data for delineation of Araucaria angustifolia crowns in the Mixed Ombrophilous Forest. The methodological procedure used classification oriented to object multi-level, through techniques of multiresolution segmentation and cognitive process of semantic net, substantiated by fuzzy rules, considering descriptors in form and texture for classification of the Araucaria crowns. Experiment was carried out in the National Forest of Irati (Parana State, Brazil). In the classification step, some statistical attributes were employed, textures (averages of sub-objects, Haralick measure) and ratio of bands, which allowed the thematic stratification of the following classes: agricultural areas, exposed ground, reforestation otPinus spp, broadleaves crowns and Araucaria crowns. The performance of discrimination of Araucaria crowns from other species was evaluated by global accuracy (0.655), Tau index and Kappa coefficient, having as reference the positioning of the trees observed in field survey.