Forearm Posture Simulation And Prosthesis Control Method Based On Electromyography Signal

Surface Electromyo Graphy(SEMG),As a physiological EMG signal can be applied in the field of rehabilitation medicine to realize the control of intelligent prosthesis and bionic hand.Currently,the pre-processed feature vector is the key point in the imitation limb control.In this paper,the whole process of myoelectric signal acquisition and preprocessing,feature extraction and action recognition,bionic model building and simulation of action posture under four action postures is studied.The method of controlling the prosthesis based on the pattern recognition of surface EMG signals used in this paper is no longer limited to the amplitude of EMG signals,but the corresponding movements are controlled by the pattern recognition results.Since SEMG inevitably mixes with background noise outside the target signal during the acquisition process,this paper designs a passband 20-500 Hz bandpass filter,an analog low-pass 2nd order Butterworth stopband filter and a high-pass filter to filter it out.Accurate detection of the effective activity segment of the EMG signal is a prerequisite for real-time control of the myoelectric bionic hand.Due to the uniqueness of the TKE operator to detect the effective activity segment,this paper proposes a dual-threshold detection method with time domain variance combined with short-time energy setting to extract the activity segment with better results.For the processed EMG signals,four characteristic parameters,integrated electromyogram,root mean square,mean power frequency and median frequency,were extracted for summary analysis.The lack of accurate mathematical models leads to poor classification between features,overlapping information,and failure to reflect the essence of the signal.Therefore,the Autoregressive Models AR coefficients are selected in the paper as feature values to be fed into the classifier for recognition experiments.The experimental results demonstrate the effectiveness of the proposed method.In this paper,the best decision function is determined by improving the standard support vector machine algorithm by replacing the first order with a second-order relaxation variable,so that the nonlinear samples are transformed into linear samples.The particle swarm optimization algorithm is also used to optimize the penalty parameters and kernel functions that affect the classification effect of support vector machines.The results are compared with the BP neural network classification results to verify the effectiveness of optimizing support vector machine parameters with particle swarm.The experimental results demonstrate that the method improves 6.3% over the standard SVM classification.Finally,a multi-degree-of-freedom bionic model is built using Solid Works to embed the proposed preprocessing,feature extraction and classifier methods.The bionic hand is imported into the virtual dynamics software Adams and MATLAB to establish a joint simulation,and the classification recognition results are converted into commands to control the bionic hand joint movements to complete the posture simulation.