Hand Motion Coding Based On Surface Myoelectric Signal

In an era of increasingly advanced technology,people hope that the machine will expand its capabilities as much as possible without abandoning its subjective control.Therefore,human-computer interaction technology came into being.The surface Electromyograph signal(sEMG)is a kind of bio-electrical signal of the human body,which contains abundant motion information of the human body.Therefore,it becomes possible to decode the motion of the human by appropriately processing the myoelectric signal.This paper mainly discusses the encoding of the right arm hand movement during the operation of the mechanical handle based on the EMG signal,and identifies the operator's motion by encoding the specific characteristics of the EMG signal.The existing pattern recognition of EMG signals simply solves the problem of motion classification,and this paper deals with the problem of motion classification with an interpretable feature level.The main contents of this paper are as follows:The simulation experiment of the mechanical handle was designed and implemented.The surface EMG signal of the right forearm and the trajectory data of the handle were recorded simultaneously during the experiment.According to the data of the handle,the EMG signal is segmented in the action mode segment,and the time domain,frequency domain and entropy characteristics are extracted after filtering.Then,five feature selection methods are adopted respectively and the effective features corresponding to the optimal method are selected based on SVM.Finally,based on the selected effective features,the decision tree and the random forest are respectively used to construct the classification treemodel.Considering the classification effect and implementation simplicity,we choose the classified tree which is constructed according to the theoretical knowledge of decision tree as the basis for the coding,and the thresholds of each node feature of the classification tree are divided,so that each path of the tree can be encoded.Experiments show the feasibility of the proposed coding method and demonstrate the effectiveness of the coding through cross-validation.The coding method proposed in this paper is interpretable at the feature level,associating encoded features with operator actions and providing a way of thinking for the feature level application of pattern recognition.