Research On Finger Gesture Classification Method Based On Surface EMG Signal

Gesture recognition refers to the automatic detection,analysis and recognition of human gestures by computer,thereby judging human intentions.The existing gesture recognition technology considers large range and single gesture,and takes into account the complexity and polymorphism of gesture movements used in daily life.Therefore,in order to recognize small gestures,12 kinds of gestures about finger movement(such as finger lifting or bending)are designed in this paper,and surface EMG signals of 12 kinds of gestures are collected.Taking surface EMG signal as the research object,12 finger gestures are classified and recognized by using hybrid kernel function support vector machine model based on vector grouping learning brainstorming algorithm.The classification accuracy of 12 gestures can reach 90%.Detailed research contents are as follows:Firstly,the design of gesture action is the premise of gesture recognition and classification.It is very meaningful to design a group of gestures which can be frequently used in daily life for applications in various fields.Therefore,in the aspect of gesture design,this paper considers the human finger movement,and designs 12 gestures according to the right hand's five fingers bending or lifting and the thumb's adduction and abduction.At the same time,the correlation between each gesture and arm muscle contraction and the placement of surface electrode patches were determined to complete the data acquisition of surface EMG signals of 12 finger gestures.Then,considering that the collected sEMG signal is easily mixed with noise interference,the Butterworth filter and 50 Hz notch filter are used to denoise the signal.In view of the importance of gesture recognition and classification,the principal component analysis method is used to reduce the dimension of the four gesture action features extracted to form a joint feature signal,which improves the model's gesture action feature signal.Recognition degree.Finally,considering the importance of the design of classification model for gesture classification and recognition,this paper uses directed acyclic support vector machine to construct multi-class classifier.Aiming at the problems of learning performance and generalization performance of single-core function in support vector machine,a new brainstorming algorithm based on vector grouping learning is proposed to optimize the parameters of hybrid kernel function SVM.Furthermore,the effects of polynomial kernel function,RBF kernel function and hybrid kernel function on the classification accuracy of the model are compared in detail.The experimental results show that the SVM model optimized by the vector grouping learning brainstorming algorithm has higher recognition accuracy and stronger stability for the collected data sets.