Identification of Gesture Based on Combination of Raw sEMG and sEMG Envelope Using Supervised Learning and Univariate Feature Selection

In this paper, we propose a novel study for gesture identification using surface electromyography (sEMG) signal, and the raw sEMG signal and the sEMG envelope signal are collected by the sensor at the same time. An efficient method of gesture identification based on the combination of two signals using supervised learning and univariate feature selection is implemented. In previous research techniques, researchers tend to use the raw sEMG signal and extract several constant features for classification, which inevitably causes a result of ignoring individual differences. Our experiment shows that both the optimal feature set and redundant feature set are not same for different subjects. In order to address this problem, we extract all the common features from two signals, up to 76 features, most of which has been established as the common EMG-based gesture index. In addition, extracting too many features in an application can reduce operational efficiency, so we apply for feature selection to get the optimal feature set and decrease the number of extracting feature. As a result, the combination of two signals is better than using a single signal. The feature selection can be used to select optimal feature set from all features to achieve the best classification performance for each subject. The experimental results demonstrate that the proposed method achieves the performance with the highest accuracy of 95% for identifying up to nine gestures only using two sensors. Finally, we develop a real-time intelligent sEMG-driven bionic hand system by using the proposed method.