Fuzzy neural nets with non-symmetric pi membership functions and applications in signal processing and image analysis

Interpolation, estimation and classification, widely used in signal process ing and image analysis, can be considered as problems of optimization. Diff erent systems could be used; some are based on known numerical data, and th e others, on expert rules. In general, they have difficulty to integrate bo th the knowledge of experts and that implied in known numerical training sa mples. In the present paper, we propose to use neural fuzzy systems with no n-symmetric pi membership functions. A new global optimization criterion an d the learning algorithm are also presented. Experimental results of applic ations to interpolation, estimation and classification problems are reporte d. The comparison with other methods shows a better behavior of such system s. Non-symmetric pi membership function gives a more general model of fuzzy rules, improving the precision of neural fuzzy system and assuring a good convergence in learning. The neural fuzzy system using non-symmetric pi mem bership functions allows integrating both the knowledge of experts and that implied in numerical training samples. (C) 2000 Elsevier Science B.V. All rights reserved.