The Study Of EEG Classification Algorithms Based On Multi-Branch 3D Convolutional Neural Network And The Design Of BCI System

In collaborative mechanical arm and the rehabilitation of the mechanical arm human-machine collaborative control research field,due to the force feedback signal control and electromyographic signal control method has their limitation respectively,it has been hard to make the user or patient with the loss of sports ability manipulate different mechanical arm or exoskeleton on different body parts,according to the thought switch or exoskeleton robot motion state to complete a series of complex movement.In order to solve this kind of problem,this study studied the electroencephalo-graph signal identification algorithm and Brain-Computer Interface(BCI)online control method.In this study,the problem of controlling robot or exoskeleton to complete complex movement with regard to the user was firstly transformed into a two-stage problem,which is composed of motor imagery(MI)identification of different body parts and motor imagery at different motion stages.Through the identification of the user's electroencephalo-graph signals in different motion stages,the user can successfully call the movement program of different body parts only through the electroencephalo-graph signals,and can switch the movement phase of the movement program timely and effectively and complete a series of complex movements according to the desire.On the basis of the method mentioned above,this study studied motor imagery classification algorithm based on three dimensional convolution neural network algorithm and the authoritative published datasets in the field of EEG signal identification.establishing a corresponding EEG signals identification model and off-line classification strategy.On this basis,this study evaluated the performance and effectiveness of the proposed algorithm by using off-line identification strategy.Experimental evaluation reveals that the proposed motor imagery of different body parts classification framework reaches state-of-the-art classification kappa value level and significantly outperforms other algorithms by 50%decrease in standard deviation of different subjects,which shows good performance and excellent robustness on different subjects.Moreover,the multi-branch structure exhibits its low latency and practicality.In addition,the algorithm proposed in this study has also achieved excellent classification performance for the classification of different action stages,and can effectively solve the problem of four motion states classification in a better real-time manner.On the basis of the research mentioned above,this study also proposes an online sampling identification strategy for EEG signal identification and an online BCI interface control system for collaborative robots.Through the identification and analysis of the EEG signals of the user,the system can identify the motion intention of different body parts and the motion intention of different stages of the movement with a low latency and a high accuracy.In the online identification of movement intentions of different body parts,the system achieved 75.593%activation rate of the movement of different body parts and an average identification delay latency value of 0.011s.In the online classification of motion intention in different stages of movement,the system achieved 83.25%identification accuracy in different stages of movement and 0.011s identification delay value.Experimental results show that the system can effectively solve the problem of using EEG signals to manipulate different robotic arms or exoskeleton devices located in different limb parts,and can also switch the motion state of robot or exoskeleton as desiring and complete a series of complex movements.