Research On High-Density Myoelectric Control Method Based On Pattern Recognition

Arbitrary human body movements are fulfilled by the coordination of multiple groups of muscles under the control of central nervous system.Electromyography(EMG)is an important kind of physiological signal that accompanies with muscular activities,it’s able to characterize the user’s real-time movement status and intention to a certain extent.On this physiological basis,the technology of collecting surface EMG(sEMG)by placing electrodes on the skin and decoding its movements information to achieve myoelectric control has become a hot research topic.In terms of control strategy,the myoelectric control evolved from the initial one degree-of-freedom(DOF)"on-off"control and proportional control to multi-DOF pattern recognition control that can achieve complex movements control,gradually approaching the human body’s movements control mechanism.With the development of electronic science and technology in recent years,the myoelectric control method based on pattern recognition has made leaps and bounds.More and more researches were devoted to exploring methods to improve the performance of myoelectric pattern recognition.However,most of these studies were limited to the laboratory environment.The myoelectric control would be challenged by many factors in the real-life application,the robustness of algorithms was usually difficult to be guaranteed,which greatly limits the development and application of this technology.The research work in this dissertation takes the myoelectric pattern recognition of dexterous finger movements as the target,aiming at the problems existing in the myoelectric control method and its application,utilizing flexible electrode arrays to collect high-density sEMG(HD-sEMG)in order to explore the novel myoelectric pattern recognition method from the perspective of mining into the HD-sEMG spatial information.The performance of the proposed methods was verified through experiments.The research work and its results can be summarized as follow:(1)Research on post-processing method of myoelectric pattern recognition based on movement pattern transition detection.This study was based on the assumption that the movement pattern would not be transient in a short period of time.The pattern distance index was designed for the movement information expressed in the "image" of the HD-sEMG features,and adopted to detect the transitions of movement pattern,thereby optimizing the conventional multi-window joint decision post-processing method,improving the robustness of myoelectric pattern recognition.The algorithm was verified in the cross-level testing scheme and the nonstop testing scheme for classification of 12 finger movements.Compared with the results of the raw classifier without post-processing,the recognition accuracies of the proposed method improved 7.33%and 10.91%respectively in two testing schemes,and furthermore,significantly higher than other conventional post-processing algorithms(p<0.05).The experimental results confirmed the robustness and feasibility of the proposed method,especially ensuring smoothness and continuity in the output of actual control command.(2)Research on multi-DOF simultaneous control method based on deep learning.This study combined the musculoskeletal mechanics and deep learning technology to solve the problem of HD-sEMG spatial information analysis.A deep neural network model was built to decode the pattern information of all finger movements.Thereby a novel multi-DOF simultaneous control method was formed,which was an effective extension of conventional myoelectric pattern recognition to the "serial" control of each DOF.The algorithm simultaneously controlled five DOFs corresponding to extension of five fingers respectively.It was verified in a simultaneous control scenario for classification of 15 hand movements involving different combinations of finger extensions.Compared with the results of conventional simultaneous control algorithms,the classification accuracy of the proposed method improved 6.04%(p<0.05),which confirmed the effectiveness of this method.In addition,the simultaneous control method has also demonstrated its application potential in predicting untrained novel finger movements with simultaneous activation of multiple DOFs.These studies aim to improve the robustness and practicality of high-density myoelectric control,and assist it to cope with various challenges in real-life applications,thus to achieve a more natural and intuitive control system,and promote the wide applications of myoelectric control in consumer electronics,robot control and rehabilitation therapy.