Study On Performance Of Myoelectric Pattern RecognitionBased Movement Classification

As the population growth, medical advances, aging development, local wars and conflicts increased, the number of limb amputations increase rapidly. At the same time with the continuous development of prosthesis control technology, many kinds of artificial limbs are commericially available for the partial restoration of limb functions in amputee's work and live. As a result of inherent limitation of traditional prosthesis control method, the current artificial limbs have the disadvantages of single function, slow control, clumsy movement and other issues.Since myoelectric prostheses has the advantage of intuitive control and natural, myoelectric pattern recognition based prostheses control strategy becomes a hot spot in current artificial limb research field. The theory basis of this method is that the electromyography (EMG) signal contains motor nerve information of human body. The intended movement of amputees can be identified by classifying EMG signals. And then the prosthesis is controller manipulates the artificial arm to complete corresponding action according to the identified movement.Extracting features from EMG signal and using pattern recognition method to predict the user's intended movement are the core parts of myoelcrtic prostheses control strategy. To extract more information about muscle contraction state and body movement from EMG signals by selecting some efficient feature parameters is one of the most important researches in myoelectric artificial field. Since EMG signal is a weak signal and affected easily by various interference factors in real time applications, studying the effect of these factors on the performance in classifying intended movements can provide guidelines for future research.In this paper, the key technology of myoelectric pattern recognition based prosthesis control strategy was studied. And a series of work was carried out around the control strategy framework. Based on EMG signal generation mechanism and collection, signal pre-progressing and feature extraction, linear discriminant analysis based movement prediction, and action execution of prosthesis, a myoelectric pattern recognition based prosthesis control method framework is developed to realize the artificial limb system. This thesis is organized as follows:Firstly, a control strategy framework for myoelectric pattern recognition based prosthesis was constructed. The myoelectric prosthesis research progress in china and at aboard was analyzed. Based on the development level of current commercial prosthesis and the users'requirement, the control strategy framework was set up and each part was introduced in brief.Secondly, a feature selection iteration algorithm was proposed and applied for feature selection. Eight time-domain features and Auto-Regression (AR) model coefficients with different order number were extracted from EMG signals. A pattern recognition method-linear discriminant analysis was used as a classifier for prediction of intended movements. The classification performance of time-domain features, AR model coefficients and the combination of two types of features were examined.Thirdly, we also investigated the effect of two real-time factors-muscle contraction force and random noise on classification accuracy through a simulation setting and then the results from simulation study were proved using those from actual experiments.Fourthly, besides the able-bodied subjects, the subjects with arm amputations were also included in this work. Only able-bodied people were selected as subjects in most of the previous studies. Amputees are the prosthesis users in practice. There are great physiological differences between the two types of groups.Finally, we give the results of classification performance, making a conclusion and propose the future research directions in this field.