| Brain-Computer Interface(BCI)is a direct connection between the human brain and external devices,which does not rely on the body's peripheral nervous system and muscular system.The intentions generated by the brain will be directly decoded and transmitted to the external devices by using BCI systems.Motor imagery electroencephalogram(EEG)signal is generated by imagination,which represents the attending moving gestures of the subject.The subject can directly control devices via his imagination by using BCI systems based on motor imagery EEG signals.BCI systems can not only help people with motor disabilities who have normal brain function,but also can be used to assist soldiers on the battlefield.However,there are problems in the practical application of BCI systems,such as lack of enough motor imagery EEG signals,low recognition accuracy of complex systems,and low real-time transmission rate.In order to realize the orientation movement and real-time monitoring of the robot system,four types experimental paradigms based on motor imagery EEG signals are designed,using the traditional time-frequency and spatial domain classification methods and the deep learning classification method.The main research outline of this paper is as follows.For the traditional method,the Wavelet Packet Decomposition(WPD)and Common Spatial Patterns(CSP)are used for the time-frequency domain and spatial domain feature extraction.To commence,the pre-processed EEG signals are decomposed into eight sub-bands by WPD,and then the CSP filters are used to extract features from the sub-bands.In order to test the feature extraction abilities of this method,an experiment is designed by using the four-class motor imagery EEG datasets of the 4th BCI competition.Targeting at improving the accuracy of multi-classification problems of EEG signals,one algorithm based on the combination of a signal to difference module,spectrogram module,signal image module and convolutional module is proposed.The signal to difference module performs multi-order differential operations on raw EEG signals to obtain its incremental representation which depicts the fluctuation features of EEG signals.Then,this representation is converted to images by spectrogram module,signal image module and convolutional module using dynamic learning parameters rather than static transformation.And pre-trained convolutional networks are applied to extract features and classify them automatically.The classification results show that our method improves the classification performance by up to8.1% when compared to recent researches.This method achieved 99.8% accuracy in two-class classification problems,92.8% accuracy in three-class classification problems and86.7% accuracy in five-class classification problems,which indicates that our signal to difference module has an important effect on EEG classification problem.At the end,an online BCI controlling system is designed based on the two feature extraction methods.The subject can control the robot to turn left,turn right,go forward,stop and turn on the camera by imagining the left hand,right hand,foot,tougue movement and closing his eyes.The results show that the subject can control the robot's movements through the BCI system.Additionaly,the traditional algorithom obtained an average recoginition rate of 69.3%,while the deep learning algorithm achieved an average accuracy of 81.9%,which proves the feasibility of the methods designed in this paper.The results of this paper provides a new control method for online motor imagery EEG signals BCI systems,which can be further developed as external communication devices and supplementary devices for people with disabilities or operator in complex situations as battlefiels. |
PDF Full Text Request