Classification of EMG signals using wavelet based autoregressive models and neural networks to control prothesis-bionic hand

This work has aimed to contribute to the prothesis-bionic hand studies. Four hundred eighty signals used in this work correspond to position of adduction motion of thumb, flexion motion of thumb, abduction motion of fingers were collected by surface electrodes. Eight healthy has participated for collecting by surface electromyogram (SEMG). The wavelet based autoregressive models of collected signals are used as feature vector for artifical neural networks. Feed forward and back propagation network, radial basis network and linear vector quantization network are used for classification in this work. One hundred twenty samples of 160 samples collected correspond to all motion are used for training cluster and as for 40 samples for testing cluster. As a result maximum accuracy rate has occured as % 90 for feed forward and back propagation network, % 92 for radial basis network and % 75,5 for learning vector quantization network.