A new fuzzy approach for pattern recognition with application to EMG classification

A fuzzy logic system with center average defuzzifier, product-inference rule, nonsingleton fuzzifier and Gauss membership function is discussed. The fuzzy sets are initially defined by the cluster parameters from the Basic ISO-DATA algorithm on input space. The system is then trained via back error propagation algorithm so that the fuzzy sets are fine-tuned. The system is applied to functional EMG classification and compared with its ANN counterpart. It is superior to the latter in at least three points: higher recognition rate; insensitive to over-training; and more consistent outputs thus having higher reliability.