Research On Hand Movement Recognition Method Based On Surface EMG Signal

The physiological electrical signal of human body is the direct reaction of human's behavioral intention.Through the analysis and interpretation of human physiological electrical signals,the machine can effectively recognize the theme consciousness of people.The surface electromyography is a kind of physiological electrical signal,which contains information about muscle state and human action intention.In order to control the peripheral equipment more effectively for the disabled,the classification and identification of the action of the opponent in this project were studied in depth based on the surface emg signals(sEMG).This project extended gesture recognition technology to the application of intelligent prosthetics,which could not only assist disabled people in rehabilitation training,but also create "phantom limb feeling" for people with disabilities.Therefore,the technique of hand motion recognition based on sEMG has important medical application value.The main research contents and innovation points of this project include:1.The degree of relative degree of each hand movement and muscle contraction position was determined to determine the position of electromyographic electrodes in the muscle.12 sEMG signals of the hand movements in daily life were collected successfully,and the sEMG signals were collected.2.A new method based on sample entropy and empirical mode decomposition was used to deal with the noise during the acquisition process.Considering the effective feature extraction method of the importance of motion feature extraction,the Fourier transform method was used to convert the time domain signal as spectrum diagram for more effectively identify characteristics of hand movements.3.The influence of different network structures on the precision of gesture recognition was analyzed in detail,and the structure of CNN convolutional neural network was constructed.By analyzing the structure of two-layer deep neural network,this project proved that the network structure containing two convolutional layers had higher accuracy and shorter training time for 12 hand movements.By adding random data sets,the whole system obtained better robustness.