Comparação de classificadores supervisionados na discriminação de áreas cafeeiras em Campos Gerais - Minas Gerais

Abstract

The use of remote sensing techniques represents a significant advance for the coffee crop data, mainly to complement the currently techniques that have been used. In this context, this study aimed to map coffee areas in high resolution images using object-oriented images analyses methods, with k nearest neighbor (KNN) and support vector machine (SVM) algorithm, and pixel-by-pixel methods, using maximum likelihood (Maxver) algorithm. The study area was mapped using two classes: ‘coffee’ and ‘other uses’. We performed the mappings accuracy analysis using reference map and it was found that the pixel by pixel classification with maximum likelihood algorithm has the best results, with kappa value of 0.78 and 94.61% of accuracy. In this study, we concluded that the pixel by pixel method of Maxver algorithm seems more efficient to discriminate coffee areas when considering only two types of land use, coffee and no coffee, in high resolution images.