Análise comparativa dos classificadores paramétricos usando imagem IKONOS 2 na Floresta de Inhamacari-Machipanda

Abstract

The quality of the results of the automatic classification process is a concern for every photo interpreter. The joint evaluation of the classification methods aims to choose better methods in order to improve the quality of the results. The present work aims to evaluate the performance of the parametric classifiers from the aerial image of the Ikonos 2 satellite for mapping the plantations of Eucalyptus spp. and Pinus spp .. For the classification first, data were collected in the study area of ​​about 250 hectares, with the aid of GPS, the coordinates of the points were marked and the area was characterized. In the computer, spectral signatures were collected for each class in the field, the separation of signatures was evaluated and then the image was classified using the three parametric classifiers present in ERDAS 9.1. From the error matrix of each method, the global accuracy and kappa indexes were determined. The Maximum Likelihood, Mahalanobis Distance and Minimum Distance methods had an order of 56.41, 33.33 and 30.77% of overall accuracy and 0.55, 0.32 and 0.29 of kappa index. Based on the values ​​of the kappa index, the Maximum Likelihood method had a good result while the Mahalanobis Distance and Minimum Distance methods had reasonable results. Through the Z test at 95% of the confidence level, the methods were compared. In the digital classification process, six classes of land use and cover were produced: Eucalyptus spp - Young (A), Eucalyptus spp-Adult (B), Pinus spp-Young (C), Pinus spp-Adult (D), Soil exposed (E) and Others (F). According to the Z test at 95% confidence level, classification using the Maxver method proved to be the best method for separating Eucaliptus and Pinus stands. (TRADUÇÃO NOSSA)