Convolutional neural network combined with subdomain adaptation for low sampling rate EMG-based gesture recognition
A new recognition method was proposed to improve the performance of low sampling rate electromyography(EMG)-based gesture recognition.The information of the pre-processed low sampling rate EMG signal was extended by an information extension layer,and the representation of key features was enhanced.In the feature extraction network,domain invariant features in the source and target domains were extracted by the subdomain adaptation network,then the domain invariant features were classified.The proposed method was evaluated using the DB1 and DB5 sub-databases of the NinaPro database.Experimental results showed that the proposed method recognized 53 and 52 gestures with the highest accuracy of 90.89%(DB1),89.90%(DB5)and 82.01%(DB1),77.07%(DB5),respectively.The effects of factors on low sampling rate EMG-based gesture recognition are reduced by the proposed method,factors that include electrode shift,muscle fatigue,changes in skin impedance,and the relative movement of the muscle relative to the electrodes.
万方数据
维普
