Research Of Human-computer Interaction Technology Based On Asyncasynchronous EEG

Brain-computer interface(BCI)is a research hotspot in human-computer interaction.It can establish a direct communication channel between the brain and external devices without relying on conventional neural channels.Since the university of Graz designed the brain-computer interface based on motion imagination,a large number of researchers designed BCI system based on EEG.However,the focus of the research is mainly on classification accuracy,interaction model of BCI in complex tasks is not well studied,and the research on asynchronous BCI is insufficient.In EEG-based speller,the main method is English speller on P300 or SSVEP,and there are few researches on Chinese speller.Focusing on asynchronous interaction model of EEG,establishment of the model,the collection of EEG signals,the classification of deep learning and the design of Chinese speller method be researched.The main work is as follows:1.A human-computer interaction simulation model based on EEG has been Established,and the relationship between single classification dimension,classification accuracy and task target number on interaction efficiency has been studied.On the basis of this,a complex task interaction method of binary classification based on EEG was proposed.Through simulation,the accuracy was increased by 7.7% and the execution time was shortened by 40% when 16 target tasks were completed.2.A template matching method based on empirical mode decomposition(EMD)was proposed,this method can detect and remove ocular artifact interference in EEG signals in an adaptive way.The results showed that the artifact recognition rate of this method reached 95%,and the EEG signal still retained valid features.3.A variable time-window semi-supervised deep learning classifier with full frequency input is proposed.For solve the problem of overfitting on deep learning models,input data structures suitable for deep learning and data enhancement for limited samples has been designed.The classification results of the BCI competition and the motor imagery data in our laboratory showed that the accuracy was improved to 87.6%,which was 6% higher than the other classification.4.A portable wireless acquisition device has been designed,in which EEG signals are collected by dry electrodes and hemodynamic signals are collected by near-infrared sensors.Devices has miniaturization,wearable and extensibility.Experiments show that the device can effectively collect brain electrical signals and hemodynamic signals.An asynchronous switch was designed using the device,and the false positive probability was reduced to 1.75% by ROC curve analysis and threshold modification.5.An asynchronous BCI Chinese speller method based on motor imagery has been designed.In this method,the system is started asynchronously,and the selected interactively by binary classification.Complex choices are reduced to binary classification.The input speed can reach 3.3 seconds per character by associative phrase input method.In this paper a multi-task human-computer interaction model has been established based on EEG.A complex task interaction method based on binary classification has been proposed.A deep learning classifier with variable time window has been designed,this classifier can improve the classification accuracy of motor imagery.An asynchronous Chinese spelling system based on EEG has designed that can realize fast Chinese input.