Research On Hand Gesture Recognition Methods Based On Few Channels Of Surface Electromyography Signals

The hand is an essential organ of the human.Hence,the hand missing caused by disease or accident will inconvenience the amputees.The flexible and stable prosthesis can help hand amputees return to daily life.The surface electromyography(s EMG)signal is a physiological electrical signal that reflects muscle activity tendency.The s EMG is hurt-free and easy to acquire,making it an ideal signal for prosthetic hand control.In the field of s EMG-based hand gesture recognition,researchers usually improve the number of recognized gestures or the recognition accuracy by increasing the number of s EMG channels.However,with the increase of the channels,the acquisition or method complexity and the data volume are also growing.Moreover,the redundant s EMG channels are unsuitable for hand amputees with few residual muscles.Hence,recognizing hand gestures based on few-channel s EMG signals remains a problem worth investigating.The main research and innovation of this thesis are as follows.(1)Hand gesture recognition based on time-domain features and neural networks.Acquiring the few-channel s EMG is simple and suitable for amputees with few residual muscles.However,the less dimension of s EMG makes gesture recognition difficult.This thesis investigates feature extraction and classification algorithms for hand gesture recognition by few-channel s EMG signals.For feature extraction methods,the timedomain,frequency-domain,and time-frequency-domain features are compared,and the results indicate that the time-domain method has the best real-time performance.For classifiers,the backpropagation neural network,support vector machine,and adaptive boosting classification algorithms are compared.The results show that the neural network has the best classification performance.The final results indicate that the recognition accuracy and computation complexity of the time-domain features and neural network is superior to other combinations.(2)Hand gesture recognition based on missing s EMG signals.The external interference may lead to s EMG missing,which affects hand gesture recognition.The commonly used signal-completion methods may increase the recognition delay,and the complemented signals may not match the original,which degrades signal processing.This thesis proposes a feature split reorganization strategy to fully use the complete signals to build different feature sets.Furthermore,the weight-based multiple neural network voting method is proposed to classify the established feature sets.The proposed methods are verified by s EMG with different missing ratios.The results indicate that the performance of the proposed methods is similar to others when the signal is complete,and the accuracy is higher when the missing signals exist.(3)Hand gesture recognition based on spiking neural network.Compared with traditional computers,the computational speed of spiking neural network is faster while the power consumption is lower.Hence,the spiking neural network is suitable for the s EMG-based prosthesis.Currently,the commonly used s EMG encoders may lead to information loss,and network decoders may degrade the performance.This thesis proposes a smoothed frequency-domain decomposition encoder to fully use s EMG amplitude information.This thesis also proposes the network efferent energy decoder and employs it as the training target to accelerate the training speed.The results indicate that the proposed methods speed up the training process and that gesture recognition accuracy is higher than traditional methods.This thesis compares different feature extraction and classification methods and selects the combination with the best performance.This thesis proposes the feature split reorganization strategy and a weight-based multiple neural network voting method to recognize the missing s EMG signals.This thesis also proposes a frequency-domain decomposition encoder and network efferent energy decoder to recognize s EMG signals by spiking neural networks.In summary,this thesis improves the classification performance of the gesture recognition system based on few-channel s EMG signals,which is beneficial to improving the performance of the s EMG-based prosthesis.