AVALIAÇÃO DOS VALORES DE ERRO DO MODELO LINEAR DE MISTURA ESPECTRAL EM IMAGENS ETM+/LANDSAT 7 A PARTIR DE REAMOSTRAGENS PELO VIZINHO MAIS PRÓXIMO E CONVOLUÇÃO CÚBICA:

This work compared the influence of two resamplings on orbital images using the linear spectral mixture model. The ETM+/Landsat 7 scene, originally available by the method of the nearest neighborhood, had it resampling changed to cubic convolution, prompting whether this change, in the face of changing values ​​of digital numbers, would influence the classification of images. 30 random samples and 30 manual samples were extracted from the transition areas of the fractions in the error images (B3, B4 and B5) of each resulting model, and the paired Student's t test for means was applied. The statistical results proved that there is not enough evidence, at a significance level of 5%, that the average of the error values ​​of the images generated by the two resampling methods in the linear spectral mixture model are different, indicating that the application of the model and the analysis of its fractions in future classifications will not be influenced using this methodology. Furthermore, the supervised classification of images and fractions, for both resamplings, found that through the confusion matrix, with an average of  99% of global accuracy, the classifications are practically identical, legitimizing that the application of different resamplings, through this methodology, did not influence the final cartography.