Research On Hand Movement Pattern Recognition Method Based On SEMG Signals

With the advent of the ageing society,the number of patients with hemiplegia caused by stroke has increased dramatically,and patients need physical rehabilitation training to regain their hand movement ability.The rehabilitation training related to hand exercise needs to be performed repeatedly for a long time,but the number of rehabilitation therapists is limited and cannot meet the clinical needs.And researches have shown that the Surface Electromyography(sEMG)signals should be involved in rehabilitation training,which shows the patients' active motor intent.And that can induce and enhance beneficial neural plasticity,and promote motor function recovery.Therefore,hand rehabilitation robots based on myoelectric control have become a research hotspot.Firstly,this paper analyzes the mechanism of sEMG signals generation and its relationship with neural plasticity and hand motor function rehabilitation.A hand autonomous rehabilitation method based on sEMG signals is proposed.It uses the patient's active consciousness for rehabilitation training,and promotes the rehabilitation of hand motor function through exercise re-learning and neuromuscular electrical stimulation.Secondly,aiming at the characteristics of weak,random,non-stationary and high degree of aliasing of sEMG signals,a blind source separation algorithm based on improved Artificial Bee Colony(ABC)optimization is adopted to de-alias the acquired sEMG signals.At the same time,to overcome the problem that the traditional sEMG signals characteristics can't fully characterize the hand movement mode,a combined feature extraction method is proposed,and the Binary Tree Support Vector Machine(BT-SVM)is used to identify the six common hand motions and the recognition rate reaches 93.33%.Finally,to overcome the problem of the poor robustness of the traditional hand motion pattern recognition system,a hand motion pattern recognition method based on Convolutional Neural Network(CNN)was adopted.The time-series high-density sEMG signal is converted into a myoelectric intensity map by a time window division strategy.Then,the CNN is designed according to the characteristics of the myoelectric intensity map.Meanwhile,the influence of time window size on the recognition accuracy and real-time performance of the hand movement mode is compared.Finally,the delay is controlled at 40 ms,and the average recognition accuracy is 98.05%.