Research On Classification Algorithm Of SEMG For Gesture Recognition

Surface Electromyography(sEMG)is a bioelectric signal that reflects muscle contraction.By analyzing sEMG signal,the movement information of the limb can be obtained.sEMG is widely used in prosthetic limb control,auxiliary diagnosis,humancomputer interaction and other fields due to its convenient collection and good bionicity.In the study of using sEMG signals to analyze human motion intentions,it mainly involves data preprocessing,feature extraction and model classification.Among them,feature extraction is an extremely critical part,but this often involves a lot of manually designed features and professional domain knowledge,so the experimenter will spend time and energy on feature extraction.In order to solve this problem,the researchers borrowed the successful experience of deep learning in image classification,and applied the deep learning method that can automatically extract features to the gesture recognition problem based on sEMG.Experimental results show that this method can significantly improve the classification accuracy of EMG signals.Based on previous research,this paper designs a dual-stream network model according to the characteristics of sEMG signals.This model can automatically extract the temporal-spatial feature of the sEMG signal.First,multi-frame EMG signals are merged into grayscale images,and Convolutional Neural Networks(CNN)are used to extract high-level abstract spatial features of grayscale images.Secondly,the sEMG signal is a type of time-series signal.There is a lot of temporal information inside the signal,and a Long Short-Term Memory(LSTM)can be used to learn the temporal features between the signals.After that,the spatial features and temporal features from surface EMG signals are fused into temporal-spatial features.By using the parallel structure of CNN and LSTM,the temporal-spatial features inside the sEMG signal can be effectively learned automatically.In addition,this article also discusses two other ways of combining CNN and LSTM,and through experiment verification,they also obtain better classification results.Finally,this paper uses the electromyograph self-developed by the team as the sEMG acquisition device,and collects five kinds of gesture data of 8 volunteers to construct the Elonxi DB dataset.The three models designed in this paper will be combined with the classic traditional classification methods and Based on CNN method for comparative experiments.In addition,in order to verify the generalization ability of the dual-stream network model,the public data set Nina Pro DB1 was used in the experimental part of this article.Experimental results show that,compared with CNN,the network structure proposed in this paper can extract the spatiotemporal features of sEMG signals,which can effectively improve the accuracy of gesture recognition.