Study On Gesture Recognition System Based On Electromyographic Signals

Gesture operation has become a hot research direction in the field of humancomputer interaction.The natural gestures of the human body have very rich meanings.The different postures of the fingers,palms and arms of the human body can be combined with a large number of possibilities,and even the speed and amplitude of the gestures can be expressed More subtle information.Although there are currently some commercial gesture interaction systems,their high price,uncomfortable wearing,and large size limit their applications in practical fields such as disaster relief and military occasions.Therefore,there is an urgent need for a gesture recognition device that is comfortable and convenient to wear,inexpensive,and accurately recognizes.We studied the gesture recognition technology based on human EMG(Electromyography)signal,and develops a complete set of wearable EMG gesture control system to realize the convenient control of terminal unmanned equipment using gestures.In terms of sensor design,the flexible electromyography sensing technology was first studied,using a conductive silver glue as the electrode,flexible printed circuit board FPC as the wire,and silicon-based material PDMS as the substrate,and designed a comfortable,convenient,and inexpensive to wear,Flexible EMG sensor with good signal quality.In terms of hardware circuit design,the ADS1298 bioelectric signal processing chip of Texas Instruments is used to perform multi-channel differential calculation on the input signal,which improves the anti-interference ability of the system.The unique right leg drive circuit pins of this chip are used to limit the common mode electromagnetic noise interference on the human skin surface.In terms of software,a host computer software was written using the qt platform to realize real-time monitoring,frequency domain analysis,acquisition and real-time gesture recognition of electromyographic signals,which improves the visualization of the system.Since traditional gesture recognition algorithms are based on manual feature extraction,in order to reduce the workload of manual feature extraction and realize endto-end gesture recognition,this paper designs a set of gesture recognition algorithms based on deep learning.The system is divided into two parts: offline training and online recognition.The offline recognition part designed a convolutional neural network model based on the VGG model,and then used different hyperparameters and recurrent neural networks to conduct comparative experiments,and found the model with the smallest parameter redundancy under the guarantee of 95% offline recognition accuracy.Reduce the time for a single identification.In order to improve the accuracy of continuous gesture recognition,a "verification-prediction" method is proposed in the online recognition part.Before a single recognition is performed,a binary classification network is first used to judge the validity of the current sample,which greatly improves the accuracy of online recognition.Reduce the situation of misidentification.In order to solve the difference problem of different wearers,a transfer learning method is proposed to improve its generalization to specific individuals.UAVs and robotic arms are used to build an experimental platform for gesture control terminals.And the system's network communication,equipment wearing,and terminal control are individually debugged and jointly debugged.The EMG signal of the human forearm can be obtained and sent to the upper computer after preprocessing by the lower computer,and then the result of gesture recognition is obtained after the neural network of the upper computer is forwarded,and finally the instructions are sent to the drone and the robotic arm,Make it complete the corresponding action.Experimental results show that the system has good stability and ease of use.