The Research On Pattern Recognition Algorithms Of Motor Imagery EEG

The Brain-Computer Interface(BCI)is the communication system which can communicate with external environment through the signal produced by our brain.This system doesn’t rely on EMG and peripheral nerves.The main research of BCI is to help people to control device through their thoughts.The key technologies are extracting features and classifying the EEG.The purpose of this paper is to build the online BCI system to control mechanical arm based on using the motor imagery EEG.Firstly,choosing the motor imagery data from BCI competition 2003 and selecting the reliable data by pre-processing the data.Due to the effective features of EEG can dramatically affect the classification results,an improved method has been introduced to classify the motor imagery EEG in this paper,whose features are extracted by using discrete wavelet transform and CSP.This method combines the features of frequency domain and spatial domain can improve the recognition performance.This method depends on the ELM and SVM with particle swarm optimization.The classification results show that the proposed method is better than the traditional methods,and the ELM needs less time to classify while it can obtain the better recognition,which proves the developed method in this paper is more adaptable to brain computer interface system.Secondly,to prove the developed method is effective,the paper analyses the offline data which collected by Emotiv Epoc.The improved ELM model is proposed during the experiment.The algorithm circulates n times by defining cycle number n,and then chooses the parameters with the highest accuracy from the n cycles.The method can improve the average recognition rate.At last,we design the program based on the Matlab platform to analyze the real-time data which collected by Emotiv Epoc,and converts the classification results to control commands.The MCU can receive the control commands,and thencontrol the mechanical arm to move.Besides,the results derived from five subjects to experiment further prove the algorithm of this paper is feasible and effective.