Exploration On Novel Myoelectric Control Strategies Using Surface Emg Pattern Recognition

Arbitrary body movements are formulated by the coordination of multiple muscles under control of the central nervous system.As an important kind of bioelectric signal,electromyogram(EMG)generated from muscular activities could be used to interpret the intention and condition of the body movements.In such physiological context,myoelectric control using the surface EMG(sEMG)recorded via sensors on the skin surface above active muscles has been developed for control purposes.This technology has been one of the hot research topics with wide applications such as prosthetic control,human-computer interaction(HCI)and rehabilitation therapy of neuromuscular diseases.In terms of control strategy,the myoelectric control has been implemented and evolved from on-off or proportional control of one DOF in early days,to multi-DOF control of complex movements in recent days.Currently,due to the revolutionary increase of the number of controllable DOFs,the control strategy based on pattern recognition has been the research focus.Given such achievements,this thesis presents a further exploration on the pattern recognition based control strategy with the purpose to develop novel control strategies and corresponding methods that promote myoelectric control close to natural motor control.The dexterous hand gestures are selected for control performance evaluation.The natural and intuitive characters of the proposed methods are demonstrated by experiments.Hence,the efforts presented in this thesis are valuable for implementing advanced myoelectric control systems that facilitate prosthetic control,human-computer interaction,and even improved stroke rehabilitation training.The primary findings and achievements of this thesis can be summarized as follows:(1)Study on the influence of muscle contraction level on EMG pattern recognition.With a high-density electrode array for sEMG recording,the relationship between EMG patterns and muscle contraction level was systematically investigated.A novel myoelectric pattern recognition method robust to different levels of muscle contraction was proposed,where the spatial features provided by the high-density electrode array were used.Furthermore,a practical control scheme was designed to improve the control practicability.The feasibility of the control scheme was proved by the experimental results.This study may also give insights about the training of myoelectric control systems and the electrode placement in implementation of myoelectric pattern recognition supporting proportional control.(2)Study on surface EMG pattern recognition using task-shared synergies.The aim of this study was to explore the application of task shared synergies in sEMG pattern recognition.Muscle synergy hypothesis had been accepted by more and more researchers as a reasonable explanation for the control mechanism underlying body movements.Nonnegative matrix factorization(NMF)was successfully applied in most muscle synergy analyses.In this study,discriminant nonnegative matrix factorization(DNMF)was utilized to extract synergies shared by a group of tasks to be discriminated.Experiments were conducted to recognize 12 hand gestures using different synergy extraction algorithms.The comparison results between DNMF and other conventional algorithms demonstrated the feasibility and effectiveness of DNMF for sEMG pattern recognition.(3)Study on multi-DOF simultaneous control using task-specific synergies.In this study,a multi-DOF myoelectric control strategy was proposed to simultaneously control movements of the five fingers.With NMF to extract task specific synergies for different states of the finger,the state of each DOF was identified by comparing distortions of the samples reconstructed with different muscle synergies specific to individual tasks/states of the finger.As a result,the final finger movement pattern can be identified through the decisions made for five individual DOFs.A group of healthy subjects were recruited to perform multi-finger movements for data acquisition.Comparing with the conventional pattern recognition method(i.e.,sequential control strategy)and the simultaneous control framework integrated with routine identification method,the proposed simultaneous control strategy was confirmed to be not only more intuitive in control mode,but also have higher control performances.Furthermore,the effectiveness of the proposed strategy in hemiplegic movement recognition was also validated by testing on the sEMG data from stroke patients.Such results indicated the application prospect of the proposed multi-DOF simultaneous control strategy in stroke rehabilitation training.