Intentions Recognition Of Eeg Signals Based On Decision Fusion In Complex Task

Brain-Computer Interface(BCI)is a communication system that does not depend on the normal output channels composed of peripheral nerves and muscles,and as a new means of human-computer interaction,BCI technology has become a hot research topic in rehabilitation engineering and biomedical engineering in recent years.The correct classification of electroencephalogram(EEG)is the key factor to determine the performance of brain-machine interface,so it is important to study the classification and recognition algorithm based on EEG.For the existing research analysis,the experimental research based on MI-BCI needs to be greatly improved and perfected,for example,the existing stage of low complexity experimental paradigm has great limitations,which cannot induce high quality EEG signals and therefore cannot accurately identify the motor imagery intention.The intention of motor imagery cannot be accurately identified.In this thesis for the above problems,research work has been carried out and the following studies have been conducted.(1)A constrained control task guided by a complex task is proposed to address problems such as the simplicity of the task in the traditional experimental paradigm.Based on this real-life scenario,the task is first conceptualized as a complex dynamic system,i.e.,a cup with a small ball for purposeful motion,in order to obtain high-quality EEG signals.(2)A classification study of EEG signals in a constrained dynamic complex object control task is carried out,using the proposed method for the study of brain network feature pattern recognition based on graph convolution combined with the characteristics of the experimental task.First,the EEG structure is established by the phase lag coefficients of multiple node EEG signals,and the time-frequency domain feature information of EEG signals is extracted as the input,then the node feature aggregation is performed by the graph convolution network to learn the spectral domain features,and finally the classification results are output by the fully connected layer.(3)In the classification research,aiming at the problem that a single classifier can not make good use of features and the adaptability of classifier,which makes it difficult to further improve the recognition accuracy,a functional brain network classification research method based on decision fusion is proposed to improve the classification accuracy of brain computer interface system.By integrating two algorithms based on LDA,this method can improve the performance of separating the false experimental features of the two classifiers,promote the selection of more likely correct decisions,and improve the accuracy of classification.