Research On Dynamic Feature Extraction Of Cognitive Activity Through Kalman Filtering

Nowadays,with the rapid development of brain science and signal processing technology,brain computer interface devices are gradually used in clinical,medical,military and gaming fields.The method of brain cognitive feature extraction is the basis of BCI development.Because the BCI technology needs to complete the man-machine interaction task in real time,the method of the brain cognitive characteristic extraction also needs a kind of EEG signal processing method with high time resolution and real-time analysis.This paper uses Kalman filtering to extract cognitive activity features.The cognitive activity feature extraction problem is divided into two sub problems: single channel EEG feature extraction and multi-channel EEG signal feature extraction.(1)In the part of the single channel EEG signal study,an adaptive Kalman filtering based single trial evoked potential extraction method and an adaptive evoked potential superiority interval separation algorithm are proposed.The algorithm uses the amplitude variation speed differences between evoked and spontaneous EEG and defines two parameters(Current Measurement Margin and the Current Estimated Amplitude).With the two parameter as the criterion,the algorithm automatically distinguishes evoked potential superiority intervals and noise superiority intervals in the EEG signals to achieve the single or a small number of evoked components accurately extracted from the single channel EEG signals.The validation experiments were conducted for the simulated evoked potentials and the practical evoked potentials.The practical evoked potential data were obtained by visual cognitive activity experiments through the Oddball paradigm.The method is applied to single trial evoked potential detection in brain computer interface system.The results show that the proposed method can effectively improve the recognition accuracy of brain computer interface system,which is as high as 92%.(2)In the part of the multi-channel EEG signal study,this paper establishes a model of speech conflict cognitive activity and proposes a time-varying brain network analysis method through Kalman filter state tracking.The method based on an existing multivariable autoregressive model uses Kalman filter to improve the coefficient estimation of multivariable autoregressive model.An effective brain network model containing frequency domain information is obtained by the causal analysis of directed transfer functions.The experimental data of auditory cognitive activity were collected by Stroop paradigm.The construction and analysis experiments of speech conflict brain network were carried out.The results show that the method extracts the structural features and frequency features of speech conflict cognitive activity and solves the information transfer of brain cognitive activities exploring the dynamic evolution process of the brain neural connections.The proposed dynamic feature extraction method of cognitive activity can well analyze the temporal,frequency-domain and spatial information of EEG signals.It provides a new approach for the modeling and feature extraction of EEG signals.