Classificação automática de cobertura vegetal em imagens aéreas e orbitais para uso em planejamento energético

Abstract

In this work, a model of neural network for energy planning and construction of energy scenarios is presented. This model permited the identification and clustering of representative pixels of water, vegetation, and anthropic impact around Tucuruí reservoir, located in the Tocantins River basin , (State of Pará, Brazil) . The database consisted in orthorectified aerial photographs and clipping of Landsat satellite’s images. Both were obtained in August 2001 and the two set of images were classified using the method of metric of the minimum distance with Matlab 7.3.0 (Matrix Laboratory – Applied mathematics software) and Arcview 3.2a (Geographic Information Systems Program). Then a competitive neural networks of Kohonen was used to classify the different areas in Matlab. This specific network allowed to map the area in n-dimension (number of entries) for a m-dimensional space (number of outputs). The results obtained with Matlab were compared with the output of Arcview classifier software. The results obtained using the neural network in Matlab and the Arcview classifier were similar, but some differences between the images in high and medium resolution were observed; these differences can be justified by the fact that the images in high spatial resolution cause many spectral variations in some features, creating ratings problems. This automatic classifier appears to be a good tool for the identification of the biomass potential for the construction of energy scenarios. The results of this work could confirm that the images in medium-resolution are the most suitable to solve the most of problems which involve the identification of land cover for energy planning.