Research On SEMG Signal Analysis And Hand Motion Recognition Method

The surface electromyography(s EMG)is a kind of bioelectrical signal generated by muscle contraction and collected by surface electrodes during muscle movement.It has the characteristics of convenient extraction and non-invasive,and has been widely used in rehabilitation medicine,Intelligent robots and other fields.Due to the rich information contained in EMG signals,a large number of hand movements can be discriminated.At the same time,the surface EMG signal is also a weak biological electrical signal,which is susceptible to noise interference,which makes the collection and feature extraction of EMG signals difficult.The research of this paper aims at the direction of hand movement classification and recognition based on surface EMG signals,and searches an algorithm with good universality,high accuracy and strong real-time performance in the analysis and feature extraction of EMG signals.It provides a feasible basis for the future bionic prosthetic hand to be promoted and used in clinic as soon as possible.For the above purpose,this paper will work from the following aspects:(1)In this paper,the mechanism and general characteristics of surface EMG signals are introduced and analyzed in detail.According to the characteristics of human forearm muscle groups,the configuration of surface electrodes is studied,and the autonomous collection process of surface EMG signals is described in detail.(2)Research on commonly used typical feature extraction methods,through analysis of typical feature advantages and disadvantages,two new feature extraction methods based on combined energy features and multivariate empirical mode decomposition(MEMD)are proposed.It caneffectively extract the effective features of signals,reduce the interference of redundant and useless signals,and make the features of electromyography signals of different gestures more obvious and easy to identify later.(3)In order to reduce the operational disaster of high dimensional features and retain the global characteristics of the signal,PCA and MDS dimensionality reduction algorithms are used in this paper for feature dimensionality reduction.By comparing the recognition effects of different feature extraction methods and different dimensionality reduction algorithms,the feature vector processed by the MDS dimensionality reduction algorithm is finally determined to be used as the input of the classifier,thus contributing to the pattern recognition of different hand movements.(4)In this paper,three supervised learning methods are used to classify emg signals,including BP neural network,linear discriminant analysis and support vector machines.In the experiment,it is found that the pattern recognition classifier is matched with the feature algorithm obtained from the previous processing,and different recognition results are obtained.In order to verify the universality of the algorithm,the classical UCI emg signal database was introduced to verify the algorithm,and good recognition results were obtained.After several experiments,the optimal algorithm matching combination is selected in this paper.The hand motion recognition algorithm proposed in this paper is applied to 8 hand motions in the autonomous collection database,and the recognition accuracy rate can reach 98.7%.For 6 hand movement databases of UCI,the verification accuracy rate reaches 98.3%;...