An Experimental Investigation of MLPNN and GRNN Classification Methods for Evaluation of Different sEMG-Extracted Features

The classification accuracy in a myoelectric control system depends on choosing the optimal features that represent surface electromyographic (sEMG) signal, and selecting robust and fast classification algorithm. In this work, eight hand motions were classified using different extracted features from sEMG signals. The results of the experiment show that the classification rate of 97.41% was achieved using wavelet coefficients as feature vector and general regression neural network (GRNN) classifier. In addition, we found that the combination of sample entropy (SampEnt), root mean square (RMS), myopulse percentage rate (MYOP), and difference absolute standard deviation value (DASDV) achieved the highest classification rate of 95.68% using multilayer perceptron neural network (MLPNN) classifier.