Fuzzy inference networks for pattern recognition.

摘要

This dissertation presents a study of fuzzy inference networks for pattern recognition problems. In this research, fuzzy neurons are defined and five types of fuzzy neurons are introduced. Three fuzzy inference models for pattern recognition systems, min-max inference model, min-sum inference model, and min-competitive inference model, are developed. Fuzzy inference networks based on the inference models and their learning algorithms are presented. The proposed fuzzy inference networks can learn fuzzy inference rules directly from training data. Two of the proposed fuzzy inference networks, Min-Max Fuzzy Inference Network and Min-Sum Fuzzy Inference Network, are applied to pattern classification problems. These two networks can learn the membership functions of all the classes and find out the soft and hard partitions according to the membership values. Another two fuzzy inference networks based on a min-competitive inference method are developed for invariant pattern recognition systems. These two Min-Competitive Fuzzy Inference Networks have been constructed for 2-D visual pattern recognition problems and have been tested with letter patterns with black and white pixel values. The learning speed of the proposed fuzzy inference networks is very fast. The structures of the proposed fuzzy inference networks are simple and they perform well when used in pattern classification and pattern recognition problems.