Human Motion Intention Recognition Based On Multi-source Neural Signals Fusion

China has been facing an increasingly severe situation of helping the elderly and the disabled since the advent of an aging society.The elderly population has a large base,rapid growth,and is showing a trend of aging and empty nesting.Elderly people with limited mobility need urgent care.Meanwhile,there are a large number of special groups suffering from motor dysfunction in our country,such as patients with spinal cord injury and stroke patients.“Helping the elderly and the disabled”has become one of the major livelihood issues that the Chinese government wants to solve.Therefore,exoskeleton robots for rehabilitation and medical application have emerged as an effective solution.At present,several products that can meet the safety requirements of exoskeleton robots are on the market.However,the synergy performance of human-exoskeleton robot systems for rehabilitation and medical application still needs to be improved.The main reason is that current research on human motion intention estimation is mainly based on physical signals such as force feedback and location tracking.Traditional physical signals have strong robustness to neurological fatigue via fusion strategies of heterogeneous sensor systems.However,compared with biological signals,there are problems for physical signals such as weakly global information,measurement lag and hence inability to achieve the usage requirements that reflect motion intentions of the wearer in a timely manner.Therefore,facing the problem of human movement intention recognition based on multi-source neural signals fusion,the study of lower-limb exoskeleton robot movement intention recognition based on multi-modal information fusion strategy is carried out.By fusing electroencephalogram(EEG)signals with surface electromyogram(s EMG)signals,accurate and global control signals are obtained with excellent real-time control performance,which is verified on the exoskeleton robot so as to provide a continuous and real-time control information source for the lowerextremity exoskeleton robot.Firstly,an adaptive sliding mode control strategy based on the mapping relationship between s EMG signals and joint torques is proposed,and the linear discriminant analysis method is used to verify the feasibility of s EMG-based exoskeleton robot control which therefore makes up for the defect of controlling the exoskeleton robot based on physical sensing signals that it cannot feed intentions of the wearer back in time.The proposed method using the radial basis function neural network has better tracking performance,higher robustness and resistance to external interference than traditional proportional-differential controllers.Secondly,aiming at the issue that intention recognition accuracy based on identical s EMG features varies from individual to individual in s EMG-based motion intention decoding research,an optimization strategy for s EMG signal acquisition channels and feature selection based on a heuristic algorithm is proposed.The adopted Relief F heuristic algorithm is based on classification performance.By changing the feature weights in the original feature set,the features are reordered and screened,and the computational complexity and amount of calculation are reduced as much as possible while ensuring high classification accuracy among different subjects.Finally,in order to make full use of the global advantages of EEG signals and the high precision of s EMG signals in the study of motion intention decoding,an EEG decoding algorithm is explored,and the control strategy of the exoskeleton robot based on EEG-s EMG fusion signals is studied.By fusing s EMG signals with EEG signals,a kind of fusion control signal that is both accurate,global and real-time is obtained,and is verified on the exoskeleton robot,thereby providing a multi-modal biological signal for the lower-extremity exoskeleton robot as the control input.The research on human motion intention recognition and exoskeleton robot control based on multi-source neural signals fusion proposed on this topic provides a more efficient,comprehensive,collaborative and accurate way for the control information source of exoskeleton robots.