In this thesis, a new pattern classification method for EMG based on soft computing techniques is proposed to help the disabled/elderly handle robotic arms of a rehabilitation system.
First, it is shown that EMG is more useful than existing input devices such as voice, laser point, keypad, etc., in view of naturality, extensibility, and applicability. Then, a new procedure is proposed to select the user-independent features. As methods to classify the pre-defined motions, a fuzzy pattern classification and a fuzzy min-max neural net(FMMNN) are designed using the selected features by the proposed procedure. It is also shown that the proposed method is evidentially less sensitive to the characteristics of users than conventional EMG classification methods such as ARMA modelling, statistical analysis, neural network, and fuzzy classification. To show the effectiveness of the proposed method, three experiments and a real application to a 6 DOF robotic arm are presented.
As results, the motions are recognized with success rates of 84 percent and 95 percent using a fuzzy pattern classification and a FMMNN, respectively.
