sEMG-Based Gesture Recognition Using GRU With Strong Robustness Against Forearm Posture

The surface Electromyographic (sEMG) based gesture recognition has been widely adopted in human–computer interaction. Traditional machine learning algorithms, such as Random Forest, SVM and KNN, have been employed for sEMG-based gesture recognition. Even though these traditional machine learning methods achieve high classification accuracy, few of reported works consider the gesture recognition robustness against different forearm postures, which often happens in real application scenario. On the other side, since the sEMG signals represent the sum of subcutaneous motor action potentials, the features of sampled sEMG under various forearm postures will be significantly different. Our experimental results show that the classification accuracy of gesture recognition using Random Forest reduces from 81% to 44%, when changing the forearm posture. In this paper, we propose a sEMG-based gesture recognition that uses recurrent neural network, specifically the Gate Recurrent Unit (GRU), to improve the robustness of gesture recognition against forearm posture. The experimental results show that the robustness against different forearm postures of our proposed gesture recognition is much stronger than that using traditional machine learning algorithms, including Random Forest, Decision Tree, SVM, KNN and Naive Bayes.