The application of an artificial neural network technique together with a feather extraction technique, viz., the wavelet transform and the wavelet package transform, for the nonstationary character of the EMG signals is described. The data reduction and preprocessing operations of the signals are performed by means of the wavelet transform and the wavelet package transform. The EMG features which extracted by wavelet transform or wavelet package transform is fed into the neural network for later pattern recognition. After training, the classifier with wavelet coefficients or wavelet package coefficients can identify eight classes of forearm movement: hand grasp, hand extension, wrist pronation, wtist supination, wrist flexion, wrist extension, forearm pronation, forearm supination with a relatively high accuracy. Experimental result shows that this method has a great potential in practical application of prothesis control.
