Research On Myoelectric Pattern Recognition Method Based On Motor Unit

Surface Electromyography(sEMG)is an electrophysiological signal generated along with muscle contraction.It is an electrical signal formed by spatiotemporal superposition of action potentials discharged by a series of activated motor units(MU)and propagated at the recording electrode placed over the skin surface,under both the filtering effect of fat,skin and other tissues and the contaminating effect of various interferences and noises.Through myoelectric pattern recognition technology,the movement patterns and intentions in multiple degrees of freedom can be well identified from the sEMG signals.For many years,researchers have carried out sufficient investigations into myoelectric pattern recognition methods based on sEMG macroscopic features.With the progress of EMG decomposition technology,it is possible to obtain the MU firing sequences and corresponding action potential waveform information,which is likely to promotes the development of novel myoelectric pattern recognition technology at a microscopic level with individual MUs.The MU is a basic functional unit in the neuromuscular system,and its activity can directly reflect the microscopic neural drive from the central nervous system to the muscle.Different movement patterns are accomplished in the way that neural commands drive a group of MUs in muscles to coordinate,and decoding movement patterns and intentions from activities of a group of activated MUs is a hotspot research direction for achieving natural and robust myoelectric control systems.At present,the MU-based methods for myoelectric pattern recognition have not been sufficiently studied yet.In this thesis,aimed at finger movement pattern recognition,a novel method for myoelectric pattern recognition is presented,relying on the assumption of MU coordination.In addition,the online operation of the proposed method is optimized by using an online sEMG decomposition algorithm,to promote the application of decoding microscopic neural drive information at an individual MU level towards advanced myoelectric control.The main findings and achievements of this thesis are summarized as follows:(1)Development of a novel method for myoelectric pattern recognition based on MU coordination.This part of research is aimed to target forearm extensor muscles for pattern classification of dexterous finger extension tasks.A 16 × 8 electrode array was used to collect high-density sEMG signals from 10 subjects performing 10 different finger movement tasks.Firstly,an algorithm for sEMG decomposition was applied to decompose the collected sEMG signals into a series of MU firing sequences and their corresponding action potential waveforms.Then,convolutional neural network(CNN)was used for classification of the decomposed MUs by characterizing their spatial waveforms spanned over all channels.According to the physiological law that multiple MUs are activated in a specific coordination way to contribute into certain muscle movement pattern and one MU may be shared in different movement patterns,a novel fuzzy weighted decision strategy was designed.This strategy was able to integrate the number,type and firing properties of the decomposed MUs,thus comprehensively assessing the contribution of individual MU to each movement pattern.The strategy produced the final movement pattern decision from a set of MU decisions,and solved the difficult problem of correlation analysis between microscopic MU activities and macroscopic movement patterns.The proposed method for myoelectric pattern recognition based on MU coordination achieved a classification accuracy of 100%,showing significant performance improvement with respect to other conventional MUbased methods and conventional methods using sEMG macroscopic features.The experimental results demonstrated the effectiveness of the proposed method.(2)Online optimization and verification of the proposed method for myoelectric pattern recognition based on MU coordination.Most of the state-of-the-art MU-based methods for myoelectric pattern recognition rely on offline sEMG decomposition algorithm,greatly limiting their practical applications.This part of the thesis focuses on the use of online sEMG decomposition algorithm to obtain the MU activity information in a real-time data processing manner.Specifically,the common offline decomposition algorithm was first applied into historical data to initialize a separation matrix,which was directly applied to the next coming data to obtain MU source signal,as well as to save computational resources.Then,possbile settings and schemes were well designed to check the reliability of considering each separated MU source signal as a decomposed MU firing sequence.With the MUs decomposed online,the proposed method for finger movement pattern recognition was conducted for performance evaluation.In addition,the impact of different window lengths on the performance of online sEMG decomposition algorithm as well as the MU-based finger movement pattern recognition was explored.The findings demonstrated the feasibility of the online decomposition algorithm for MU recognition and myoelectric pattern recognition,and provides a new solution to MU-based myoelectric control system.In this thesis,a novel method for myoelectric pattern recognition based on MU coordination is presented,and its online operation was examined and optimally customized towards actual myoelectric control applications.The efforts reported in this thesis provide a new tool for decoding movement intentions from the perspective of microscopic neural drive information at an individual MU level so as to promote advanced application of MU-based myoelectric control technology,especially in the fields of prosthetic control and rehabilitation treatment.