Evaluation of EMG Feature Extraction for Movement Control of Upper Limb Prostheses Based on Class Separation Index

Abstract

To control of the upper-limb prostheses based on surface electromyography (EMG) movement actions, the first and the most significant step is an extraction of the efficient features. In this paper, an evaluation of various existed EMG features based on time and frequency domains is proposed by using a statistical criterion method, namely RES index, the ratio of the Euclidean distance to the standard deviation. The RES index can response the distance between movement scatter groups and directly address the variation of feature in the same group. Moreover, the evaluation of EMG features based on the statistical index does not depend on the classifier types. The EMG signals recorded from ten subjects were employed with seven upper-limb movements and eight muscle positions. Fifteen features that have been widely used to classify the EMG signals were tested with three real-time window size functions including 256, 128, and 64 samples. From the experimental results, Willison amplitude (WAMP) with threshold value 0.025 volts shows the best performance in class separation compared to the other features. Waveform length (WL) and root mean square are useful augmenting features. Two efficient features, i.e., WAMP and WL, are suggested to use as a feature vector for the EMG recognition system. It will be obtained the high classification accuracy and can be reaching for the real-time control system. Moreover, the effect of window-size functions is dependent on the type of features.