Multi-joint Motion Decoding Of Upper Limbs Based On Motor Imagery EEG Signals And Its Application

Stroke is the primary factor to death and disability in the adult population in China.The main reason is that most patients do not receive timely and effective rehabilitation.The brain computer interface(BCI)technology has been developed to provide a new rehabilitation stratergy for stroke patients.The BCI technology based on motor imagery(MI)helps patients control prosthetic or exoskeleton for rehabilitation,and at the same time induce nerve repair by inducing cortical plasticity.Currently,the recognition accuracy of MI tasks is limited,and only a few simple limb movements can be identified.In order to make the control process more natural and smooth,more complex MI task recognition has become a recent research topic,and the recognition of upper limb rehabilitation movement has attracted more and more attention.The active areas relevant to complex MI tasks in the spatial cortex are very close,which makes it difficult to distinguish different MI tasks.Therefore,it is necessary to propose a more effective method to classify the electroencephalography(EEG)signals for fine movements of the upper limbs.In this work,the online detection method of the subject’s motion state and multi-joint motion decoding of upper limb are studied in depth,and an effective idle and motion state recognition algorithm and MI classification method are proposed.Finally,an upper limb multi-joint motion BCI rehabilitation system based on MI EEG signals was designed and implemented.The main research work is as follows:(1)Online detection of idle state in MI based on Movement Related Cortical Potential(MRCP)and Common Spatial Pattern(CSP).Considering the problem of real-time detection of idle and motion state in asynchronous BCI system,the traditional CSP algorithm and MRCP are combined to extract the temporal and spatial characteristics identify the motion state of the subject.Two classification models are established according to the two transition processes of idle state and motion state.In the asynchronous system,the classification model is selected adaptively according to the current motion state of the subject,so as to realize real-time and high-precision detection of the motion state of the subject.(2)Multi-joint MI decoding of upper limbs based on weighted optimization EEGNet.An EEG signal classification algorithm for fine motions of upper limbs based on weighted optimization EEGNet is proposed to solve the problem of difficulty in extracting EEG signal features of fine movements and low recognition rate.Firstly,the extracted EEG signal of specific frequency band is denoised by discrete wavelet transform,and data augmentation is performed on the original sample by adding Gaussian noise.Then the EEGNet based on deep convolution and separable convolution was used to extract and classify features.At the same time,the initial weight of EEGNet are optimized by genetic algorithm.The results show that compared with common neural network structures and traditional method,this deep learning framework can effectively avoid the problem that the classification result is greatly affected by the initial weight of the network,and improve the average recognition accuracy of the MI tasks.(3)The application of brain-computer interface based on the upper limb multi-joint MI.Research and design a set of asynchronous BCI system based on multi-joint MI of upper limbs.Through signal acquisition and signal processing module,EEG signals are collected,preprocessed,feature extracted and classified.Combined with hierarchical classification to realize six kinds of motion recognition:elbow flexion/extension,wrist pronation/supination,and hand grasp/open.Finally,according to the joint and direction information of the upper limb movement,the human–computer interaction module and the manipulator control module are used to control the robotic arm to complete the daily tasks of multi-degree of freedom.