Classificação orientada ao objeto de imagem Quickbird-2 para a identificação da Araucária

Abstract

The objective of this study was to map the crowns of Araucaria angustifolia from a Quickbird-2 image using the object-oriented classification. The study area is located into a fragment of Mixed Ombrophilous Forest located on National Forest Irati - Paraná, Brazil. Based on aerial photographs at 1:2.000 scale, it was carried out fieldwork in an area of 10 hectares, collecting informations about the isolated and clusterings of Araucaria crowns. In the image Quickbird-2 was carried out some processing, such as: fusion, geometric correction, segmentation and object-oriented classification. It was tested four methods of image fusion, and through qualitative and quantitative analysis, Gram-Schmidt and Principal Components methods were superior to the others, with correlation coefficients of 0.93 and 0.95, respectively. Geometric correction was made on the fused image, resulted from the Principal Components method, which resulted errors below one pixel, acceptable for the study. In the segmentation process were used the following scale values: 80 for segmentation, and 90 for the merge segments. In the object-oriented classification, using the composition 4R3G2B, it was tested the supervised classification processed by the K Nearest Neighbor algorithm and the classification based on logic fuzzy rules. For the two thematic classes of interest, Araucaria angustifolia and hardwoods, it was highlighted by the minimum and maximum values of the near infrared band, NDVI (band ratio) and the attribute space "area." For the thematic class "shadow" stood out the minima and maximum values of green band and the intensity (band ratio). The resulting Kappa value of 0.83 and 0.82, for the supervised classification and by rule-based classification respectively, was considered as excellent. The Quickbird-2 image was efficient for the crown mapping of Araucaria angustifolia in both tested classifications.