Study On EEG-based On-line Control Method For Upper Limb Prosthesis

Limb disability especially upper limb complete loss will bring great inconvenience to patients and seriously impair their quality of life.With the continuous improving of people's living standards,the prostheses with higher performance are required for the amputees,which are controlled naturally for having multi-degree of freedom and more intelligence.So the research on prosthesis control based on EEG,EMG,MMG,ENG and other signals in the human body,has been a hot and front topic in robotics and biomedical engineering fields,which has great significance in both theory and practice.But it is very difficult to gather and recognize the above mentioned signals,so the artificial limbs controlled by these signals have only a few degrees of freedom,poor reliability and real-time performance.Left hand,right hand,left shoulder and right shoulder motor imagery EEG signals are chosen as the source to control the full-arm artificial limb with multi-degree of freedom designed in our laboratory.And the on-line BCI system platform based on motor imagery for controlling the upper limb prosthesis is developed.Based on this,space target object localization and capturing methods in prosthesis workspace are designed and the hardware platform of prosthesis motion controller is established,thus the purpose of contolling prosthesis to grasp the target object at any position is achieved by only four-class motor imagery recognized.In accordance with the mechanical structure of the designed multi-dof artificial limb and the controlling of moving somewhere and grasping the object there,the control system architecture and the on-line operating method based on left hand,right hand,left shoulder and right shoulder motor imagery EEG are put forward and designed.The experiment scheme of collecting the EEG and the data selection are discussed for improving the quality and reliability of the collected signals.AR model power spectrum estimation and continuous wavelet transformation are used to analyze ERD/ERS distribution of EEG signals induced by four imaginary movements of left hand,right hand,left shoulder,right shoulder.The result shows that imaginary movement of left or right shoulder can also induce different ERD/ERS phenomena and the induced channels and frequency band of ERD/ERS are different between imaginary movements of hand and shoulder on the same side It indicates that the proposed four kinds of motor imagery EEG signals are classified On the basis,feature extraction methods based on wavelet analysis are deeply researched for ERD/ERS feature and phase synchronization feature.An approach combining continuous wavelet transformation and common spatial pattern is proposed to extracting ERD/ERS feature,in which the associated selection of time,channel and frequency is discussed.In order to enhance classification accuracy and realize real-time control of prosthesis,combined feature of ERD/ERS and phase synchronization is presented,and the classification results show that the combination of above two features is effective for distinguishing left-right hand and left-right shoulder motor imagery tasks.Four-class classifiers based on PNN and SVM are respectively studied in identification of the EEG signals induced during left-right hand and left-right shoulder motor imagery.According to classification principles and characteristics of two algorithms,the classification strategy is selected separately,then the corresponding classifier is designed and the involved parameters are discussed and optimized.Furthermore,the multiple classifiers fusion under two algorithms is adopted and researched to update the classification accuracyThe software of on-line prosthesis BCI system based on motor imagery EEG is designed,and the goal is reached for controlling the prosthesis to move and grasp by four imaginary movements of left-right hand and left-right shoulder is reached.The software is also used in offline acquisition and analysis of motor imagery EEG data Moreover,it allows adjusting many parameters in data collection and recognition and the adjustment is required for each subject considering individual differencesThe method with robot control principles is designed for localizing and capturing object in the multi-DOF prosthesis workspace.Firstly,the prosthesis kinematics model is set up,and a localization method is designed with laser range finder sensor and posture sensor mounted on the head and shoulder of prosthesis user.Furtherly,a solving algorithm for joint space parametric variables in the case of incomplete position and attitude information is proposed based on spatial traversal and BP neural network,and the actual application shows that it is good for giving a feasible solution of controlling prosthesis to grasp the target object with appropriate posture.