Evaluation of EMG feature extraction for hand movement recognition based on Euclidean distance and standard deviation

In EMG hand movement recognition, the first and the most important step is feature extraction. The optimal feature is important for the achievement in EMG analysis and control. In this paper, we present a statistical criterion method using the ratio between Euclidean distance and standard deviation, which can response the distance between two scatter groups and directly address the variation of feature in the same group as a selection tool to find the optimal EMG feature. Fifteen features that have been widely used to classify EMG signals were used. The optimal feature is conducted to demonstrate the validity of the proposed index. The major advantages of this method are simplicities of implementation and computation. Moreover, the results of proposed method are the same trend with classification results of the achievement classifiers in EMG recognition. From the experimental results, waveform length is the best feature comparing with the other features. Root mean square, mean absolute value, Willison amplitude, and integrated EMG are useful augmenting features for a more powerful feature vector. From these results, it demonstrates that the proposed method can be used for an EMG feature evaluation index.