Fuzzy-neural pattern recognition
M. Engel, M. Leclercq, C. Pradels · 2002
The common techniques of neural networks seem to apply perfectly to the classification and to the pattern recognition, but the learning can sometime be long and the complexity of the final network very big. An association between the neural networks and the fuzzy may seem feasible and make significantly easier the learning period and to allow also to simplify its structure. The fuzzy logic expresses the knowledge in an explicit manner by its rules of inferences, while the neural networks assimilate that knowledge in the weight during the learning, the knowledge is so here implicit. In merging these two approaches, we can introduce the connectionist techniques in the fuzzy logic operating in the best way for every underlying analogies between these two approaches. It is that last possibility that is handled in that work. We can, by that association build a neural network of simple structure where we meet the stage of the fuzzy reasoning. The work here described tries in a first time to identify a paraboloid by a simple structure in the aim to well understand the developed association in the aim to apply it subsequently to the recognition of characters.>