Motor Imagery Electrocorticography Signal Perception In Brain Computer Interaction

Brain computer interface(BCI)is an emerging human-computer interaction technology.It uses electroencephalography(EEG)signals to directly control equipment.Motor imagery(MI)refers to the subject’s direct control of devices by imagining movements.It widely uses in various fields.However,with the features of weak signal,strong interference noise,it is difficult to identify MI-EEG.For extracting features with high discrimination,the studies were conducted.1)Due to the nonlinear characteristics of EEG signals,permutation entropy(PE)was proposed to feature extraction,which retained the small signal changes.The accuracy is 90%,which was higher than the traditional entropy method.For the problem of single feature type,the WPT-DE method was proposed,which was combined wavelet packets and differential entropy.The accuracy of WPT-DE is 88%,which is improved by4% and 8% respectively compared with the single feature extraction method.2)For the problem of the heavy workload,a method based on Original-WPT-multi stream Convolutional Neural Networks(OWMCNN)was proposed.The automatic feature extraction of CNN is used to process the signal.The preprocessed signal and the WPT reconstructed signal are respectively input into CNN for secondary feature extraction and classification.The average classification accuracy is 91.7%.3)For the individual user differences affect MI-EEG BCI control efficiency,combined with OWMCNN method,a MI-EEG virtual training system is designed based on the Unity3 D open platform.It provides a low-cost MI-EEG training platform and improves the adaptability of users to the BCI system.To sum up,compared with the traditional methods,the improved MI-EEG feature extraction and classification method increase the accuracy.It provides a new idea for MIEEG data processing and the application of MI-EEG BCI technology,so as to Promote the application of BCI technology.Figure 40;Table 11;Reference 55...