Contextual dynamic neural networks learning in multispectral images classification

Various methods for integrating spatial contextual information in multispectral images classification have been developed during the last two decades. These methods have for a large part been of two main types: 1) Pre-classification neighborhood-based using spatial contextual correlation between adjacent pixels in the spectral space, 2) Postprocessing-based smoothing using the contextual correlation between adjacent pixels in the "decision" space. In this paper, an iterative contextual classification algorithm is developed. The aim of this algorithm is to use the spatial contextual correlation between adjacent pixels in order to form the data base to be used in learning a neural network classifier.