The Motor Imagination EEG Recognition Combined with Convolution Neural Network and Gated Recurrent Unit
作者单位：Chongqing Key Laboratory of Complex Systems and Bionic ControlChongqing University of Posts and Telecommunications
会议名称：《第37届中国控制会议》
会议日期：2018年
学科分类：0831[工学-生物医学工程（可授工学、理学、医学学位）] 12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 0836[工学-生物工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）]
基 金：supported by National Natural Science Foundation(NNSF)of China under Grant 61673079 Chongqing Basic Science and Frontier Technology Research Project of China under Grant cstc2016jcyjA1919
关 键 词：Motor Imagination EEG feature extraction Convolution Neural Network Gated Recurrent Unit
摘 要：For the common electroencephalogram（EEG） feature extraction methods, the temporal sequence of EEG signals is often neglected. A new model called Convolutional Gated Recurrent Neural Network that combines Convolution Neural Network（CNN） and Gated Recurrent Unit（GRU） is proposed. The model extracts the combinatorial features of the preprocessed Motor Imagination EEG by CNN, it enriches the GRU input, and then it uses GRU to extract some sequence information hidden in the EEG signals to improve the recognition accuracy of MI EEG signals. In addition, batch normalization（BN） is added between the CNN and the GRU to speed up the operation of the network, and a dropout is introduced to sparse networks’ connection to prevent over-fitting. Experiments results show that the average recognition accuracy of the EEG data collected by ourselves is 93.57%. In the open data set BCI competition IV data set 2 b, compared with the traditional CSP algorithm, it can improve the recognition rate of EEG by 2.56% on average.