Deep Learning with Attention on Hand Gesture Recognition Based on sEMG

Abstract

The human forearm Surface Electromyography signal (sEMG) is related to gesture activities, and the human body movement intention can be predicted by analyzing and identifying the forearm sEMG signal. Deep learning has been widely used in gesture recognition research because of its ability to extract deep features. On this basis, this paper introduces an attention mechanism to assign weights to different channels, so that the acquisition of the model is more dependent on some explicit channels to obtain a model with better performance. Compared with other models, the model proposed in this paper not only has fewer parameters, but the experimental accuracy rate on private datasets can reach up to 99.6%, which is comparable to some current CNN network models with good classification effects; In the case of the smaller datasets, the model can still maintain more than 95% accuracy and has good adaptability.