Feature Extraction And Classification Of Motor Imagery Eeg Signals Based On Convolutional Neural Network

The Brain Computer Interface(BCI)technology refers to not rely on normal brain nerve and musclar tissue,but on computer or other equipment to build a new information transmission circuit between the brain and the external environment,it can directly realize the information exchange between brain and external environment,which is a new way for human to communicate with the outside world.BCI can provide a new way to communicate with the outside world for people who have normal thinking but suffering from muscle damage or muscle dysfunction.Also,in military applications,entertainment and other fields BCI still has a widely application prospect.The key technology of BCI system is that after processed we can extract components of EEG signals which can characterize the subjects' thinking activity,then the characteristic is sent to the classifier as the input of the classifier,and the classification result can be converted into a control command,thus realizing the brain controls the external equipment.In order to improve the recognition rate of four kinds of motor imagery EEG system,this paper analyzes the EEG signal preprocessing,feature extraction and classification,which is based on the characteristics of motor imagery EEG signal.The main work of this paper is as follows:(1)Firstly,the classification and characteristics of EEG signal are introduced and analyzed.Then,combining Wavelet Packet Transform(WPT)algorithm and Fast Independent Component Analysis(FastICA),the EEG signals are preprocessed to filter out high frequency noise and artifacts,thus facilitating the follow-up such as the feature extraction and classification.(2)When using the deep Convolution Neural Network(CNN)to classify the EEG signals,due to the small sample size,the network weights are not adequately trained,thus resulting in low classification accuracy rate.Regarding the issue referred above,this paper innovatively combines Common Spatial Pattern(CSP)with CNN algorithm to extract and classify the EEG signals of multi-class motor imagery.Compared with the traditional EEG classification algorithm,the classification accuracy rate has been improved.(3)The small sample size of the EEG signal can not adequately train the weights of CNN neural network,and the initial value of network weights in different experiments has a great impact on the classification results.Aiming at this problem,in this paper,the CNN weights are firstly pre-trained by Genetic Algorithm(GA),and the weights are globally optimized in the solution space.Secondly,the BP algorithm is used to modify the optimal network weights found by genetic algorithm.The result shows that the proposed method can achieve better CNN training effect in the case of small sample size,which is more stable than the traditional classification method using CNN algorithm to carry out weight training.