The Research On Myoelectric Control Of Upper Limb Rehabilitation Robot Based On Pattern Recognition

The traditional therapeutic methods to hemiplegia are mostly by hand-to-hand guidance that comes from physiotherapists or by some simple automatic equipments. However, the rehabilitation training often fails because of inefficiency or patients' resistance. The upper limb rehabilitation robot controlled by electromyography signal based on pattern recognition provides an effective solution to the problem of hemiplegic patients' rehabilitation training,and has a great application prospect. Therefore, the research work of this paper mainly focus on feature extraction and classifier designing to select the feature and classifier combination of high recognition rate for the target movements, to establish a real-time myoelectric control system of upper limb rehabilitation robot, and provides theoretical basis and technical support for the clinical application of the robot.The first section builts a data acquisition system based on mechanism and traits of electromyography signal, and selects six muscles for recognition of the fourteen target movements. Then presents a new method to distinguish the beginning and ending of the target movements using integral electromyography, and improves the system's accuracy by IIR filting.The second section constructs feature vectors of electromyography signal using time domain features , fourth-order autoregressive model coefficients, maximum value of wavelet coefficients, energy value of wavelet coefficients, respecttively. Then presents a fusion method using two different features for the problem of single feature vector containing less information.The features are the time domain features with wavelet coefficients, the autoregressive model coefficients with wavelet coefficients. And analyzes the clustering of feature itself with average value and standard deviation parameters.The third section uses particle swarm optimization algorithm and Levenberg-Marquardt algorithm to optimize BP neural network for it's slow convergence and easy to fall into local optimum. Then designs three classifiers of BP neural network, support vector machine and hidden markov model to recognize the target movements. And analyzes the classification effect using average with the lowest recognition rate of the target movements, to find out that time domain features fusion with wavelet coefficients combine with support vector machine have a ideal classification effect for the target movement.Finally, on the basis of analysing the trailed upper limb rehabilitation robot from two aspects of mechanical transmission and the movements' inverse kinematics, real-time myoelectric control system is constructed, including hardware system and software system.And designs experiment to verify the effectiveness of the system.