Research On Gesture Recognition Method Based On SEMG Signal

No matter in the field of civil,medical or military,human-computer interaction has been widely used,such as intelligent prosthetic control,sports medicine,rehabilitation medicine,clinical medicine,combat command and leadership,etc.,and it has gradually become a focus of current social scientists' research.In the field of intelligent prosthesis,surface electromyography(semg)has become the most widely used control signal source due to its advantages of easy acquisition and more direct,natural and non-invasive control of prosthesis.For the present situation,it is still a problem to control the intelligent prosthesis by using the surface emg signal.First of all,there are only a few types of gesture recognition,and they are all basic gestures and wrist movements.Secondly,the off-line surface emg signal control of the intelligent prosthesis does not have real-time performance,and the delay from the issue of instructions to the completion of the movement will cause confusion to the user.Then,the traditional feature extraction method requires a large number of complex calculations,which is time-consuming and labor-intensive,resulting in low accuracy of gesture recognition and affecting the precise control of intelligent prosthesis.In view of the above problems,this paper studies the feature extraction and motion recognition of semg signals for some finger movements.The main contents of this paper are as follows:(1)this paper introduces the current situation of intelligent prosthesis at home and abroad,the research status of surface emg signal feature extraction method and recognition method,and expounds the corresponding generation mechanism,mathematical model and characteristics of surface emg signal.At the same time,the DB2 health individual data set and DB3 disability individual data set in NinaPro database used in this paper are introduced,and the data of 9 kinds of finger movements are normalized,sampled and filtered.(2)feature extraction and dimensionality reduction were carried out for the preprocessed surface electromyographic signals by means of principal component analysis method,and then classification was conducted by using support vector machine algorithm.The average recognition results of all subjects in the two data sets of DB2 and DB3 were 60.753% and 45.238%,respectively.(3)the total space mode algorithm in the feature extraction of multi-channel semg,use "one to one" space model algorithm for multiple class action for feature extraction,linear discriminant analysis method is then used as a classifier to classify the corresponding gestures,and use the total space of the two kinds of improved algorithm are respectively Tikhonov regularization of space model algorithm and weighted Tikhonov regularization of space model algorithm,space pattern with the original algorithm,this paper compares and analyzes TRCSP algorithm is the highest recognition rate,DB2 data collection of all the individual average recognition rate of 78.853%,The average recognition rate of all subjects in DB3 data set is 63.492%.(4)extract the RMS of the surface emg signal data,and then design a deep convolutional neural network framework for pattern recognition of 9 gestures.The extracted signals are sent into the network,and the recognition rate is 86.71% for the data set of healthy individuals.For the data set of disabled individuals,the recognition rate is 72.06%.This chapter also compares several recognition algorithms used in this paper.The CNN algorithm has the best recognition rate but takes the longest time.Although the TRCSP algorithm has a lower recognition rate than CNN,its operation speed is fast.