Comparison of neuro-fuzzy, neural network, and maximum likelihood classifiers for land cover classification using IKONOS multispectral data

Abstract

For the comparison and evaluation of neuro-fuzzy, neural network, and maximum likelihood classifiers, a land cover classification activity was performed using multispectral IKONOS data of part of Daejeon City in Korea. For this purpose, a neuro-fuzzy program was derived from a generic model of a three-layer fuzzy perceptron. The results of the classification and method comparison show that the neuro-fuzzy classifier was the most accurate method. Thus, the neurofuzzy model is more suitable for classifying a mixed-composition area such as the natural environment of the Korean peninsula. The neuro-fuzzy classifier is superior in its suppression of classification errors for mixed land cover signatures. The classified land cover information is important when the results of the classification are integrated into a geographical information system.