Possibility-based fuzzy neural networks and their applications

A generalized fuzzy neural network has been created and implemented for this thesis. The major difference between a standard neural network and the fuzzy neural network proposed in this thesis is that the fuzzy neural network can accept a set of possibility functions as input as well as a vector of scalar values. This neural network has been used in three applications: satellite image classification, seismic straitgraphic pattern recognition, and ionogram scaling. The results show that the fuzzy neural network compares well with other neural network and image processing methods.;This thesis describes the implementation of a two-stage fuzzy neural network. The first stage of the network is fuzzy-based in that the weights of the associated nodes are implemented as possibility functions. The output of this layer is a fuzzy set. This output forms the input to a standard backpropagation-based neural network. Such a fuzzy neural network shows promise for the classification of complex feature sets because it performs better with fewer nodes and layers than a comparable performance backpropagation-based neural network.;This thesis explains the reasons for introducing the proposed neural network through several practical problems. This thesis then examines theoretical results based upon the neural network. The possibility-based neural network is self-complete and unique.