Avaliação e compação de métodos de classificação de imagens de satélites para o mapeamento de estádios de sucessão florestal

The objective of the present work was to compare the orbital images from the CBERS 2, IRS-P6 and the Quickbird satellites, for the mapping of the stadiums of ecological succession, using different methods of classification of orbital images. The study area included the native forest reserves of the Macedonia forest project, located in the cities of Bugre and Ipaba, between the parallels 19o19'00"S and 19o24'30"S and the meridians 42o27'00"W and 42o21'00"W, Minas Gerais state. The visual, per region and pixel-based classifications were used. In order to evaluate the classification accuracy of each method, a matrix of errors was generated and the Kappa index was calculated. The Z test was used in order to test the statistical significance of the difference between two Kappa indices. Altogether, the best results were the classifications obtained in the visual and region-based methods, with values of kappa higher than those for pixel-based classifications. The image resulting from the fusion of IRS and CBERS images, which were classified through the method of regions, obtained the best Kappa index, and was in the level considered as good. The problems of separation among the classes resulted in classifications with a low degree of accuracy. This can be explained by the spectral similarity present between the targets (initial, medium and advanced stadiums), the small variation among the numerical values of the pixels, the existence of overlapping among classes and the low spectral resolution of sensors.