Feature Evaluation to Reduce False Triggering in Threshold Based EMG Prosthetic Hand

The attempts to make cost effective threshold based prosthetic upper limb suffered from the problem of false triggering, the wrong motion of prosthetic hand because of randomly introduced noise in the EMG (Electromyogram) signal. EMG itself being a random signal makes the problem worst. Hardware filter works for the predictable and known noise. Kalman filter works better for unpredictable noise but at the cost of increased computational time thus reducing speed. A better selection of the feature as a classification criterion can solve this problem. If a feature can provide large interclass difference for two different motions, good immunity for the randomly introduced noise can be achieved. Simple algorithms have been used to evaluate the features rather than using Fuzzy or Artificial Neural Network in order to achieve a cost effective and less complex in computation. The EMG signal was acquired using the analog components like AD524 (Instrumentation Amplifier), Notch filter and operational Amplifier. The signals were then sampled at 2.5 KHz sampling frequency. The six features evaluated for the classifications were IEMG (Integrated EMG), Variance, ZC (Zero Crossing), WL (Waveform Length), Wilson amplitude (WA) and Slope Sign Change (SSC) were calculated and evaluated using MATLAB (7.0) Software. The results show that the features have distinct values for different motions. Based on the EMG data acquisition and processing of these signals it can be conclude that Variance and IEMG are the most effective features, for classification of motions, among the evaluated features. The chances of false triggering for opening and closing can be highly reduced as the highest ranking favorable features (Variance and IEMG) have a significant difference, in value, for two different motions, open and close.