A Fuzzy Context Neural Network Classifier for Land Cover Classification

Abstract

Land cover classification based on statistical pattern recognition technique applied to multispectral remote sensor data is one of the most often used methods of information extraction. Among various classification techniques, neural network classifier makes no strong assumptions about the form of the probability distributions and can be adjusted flexibly to the complexity of the system that they are being used to model, therefore considered to be an attractive choice. However, traditional classifiers are often referred to as point or pixel-based classifiers in that they label a pixel on the basis of its spectral properties alone. In this paper, we present a new context-sensitive neural network classifier, which take into account the spatial context information, using fuzzy method and probabilistic label relaxation. The experiment result shows that the new classifier can reduce some isolated mislabeling and improve the accuracy. The spatial coherence of the classes improved.