Classifier Design Of Hybrid Brain-computer Interface Based On EEG-NIRS

Brain Computer Interface(BCI)is an emerging field involving interdisciplinary research.It is an important branch in the field of brain science to directly recognize the physiological signals of the brain and control the external equipment by using human-computer interaction technology.This technology provides a solution for the reconstruction of motor function in patients with impaired or even paralysed motor ability,and it also has important application value and scientific significance for the exploration of brain cognitive process and the processing of intelligent information.This thesis focuses on the signal classification of hybrid BCI system based on EEG-NIRS in motor imagery(MI)task.Firstly,the characteristics of electroencephalogram(EEG)and near-infrared spectroscopy(NIRS)signals in hybrid mode BCI system were analyzed and extracted,and then convolutional neural network(CNN)and long-term memory network(LSTM)classifier were combined to achieve efficient and accurate classification of mixed mode BCI system signals.The following findings were obtained in this study:Firstly,the definition,composition and classification of BCI system were introduced in detail.The generation principle and characteristics of EEG signal and the principle and characteristics of near infrared spectroscopy were demonstraed.Secondly,the source of experimental data and the process of data preprocessing were analyzed.The common feature extraction methods of EEG signal and NIRS signal were discussed,and the feature extraction methods of NIRS signal were compared emphatically.Finally,common spatial pattern method was selected to extract EEG signal features,and first derivative method was selected to extract NIRS signal features.Finally,the commonly used classifier model in BCI field was discussed.According to the characteristics of EEG signal and NIRS signal,the CNN-LSTM fusion neural network classifier was improved and designed.Compared with the traditional classifier,it is found that the classification accuracy of the algorithm had been greatly improved.This strongly illustrated the effectiveness of the classifier fused in the full connection layer,and provided a novel idea for the hybrid BCI signal classification based on EEG-NIRS.