Sistema de inferência fuzzy geral do tipo-2 aplicado à classificação

Abstract

This work proposes the development of a new tool based on general type-2 fuzzy sets to be applied to digital classification of data. The classification problem considered here regards the identification of areas of forest in satellite images. The goal is to assist users in tasks related to monitoring forest. The developed digital classifier employs an inference mechanism called "general type-2 scaled inference" to classify pixels in images according to their vegetation cover. Such classifier is innovative because, besides using general type-2 fuzzy sets, it can use specific and generic rules base (both in a hierarchical way) to reclassify pixels that remain unclassified. Such hierarquical reclassification leads to a compact rule base (with few rules). The reason why one should use type-2 fuzzy inference is that they present better performance than their type-1 counterparts, in spite of their bigger computational cost. The carried out tests showed, for sure, that the proposed system is better than the conventional fuzzy classifier usually employed in similar applications and its performance is comparable to the statistical likelihood classifier, proving to be an alternative choice to this last one.