Research On EEG-based Emotion Recognition Technology For Passengers Of Autonomous Vehicles

With the development of microelectronics,biomedicine and BCI,monitoring physiological signal such as EEG and ECG remotely by utilizing wireless body network is gradually becoming normal.However,the long-term monitoring of the multi-channel EEG signal in the wireless network is still limited by the bandwidth of the data transmission node and the power consumption of the system.Conventional data compression methods are computationally complex and difficult to implement on sensor nodes with weak computing power.Fortunately,the front-end compression coding of compressed sensing theory is computationally simple,and the complex computation is mainly focused on the back-end reconstruction.This makes compressed sensing suitable for the front-end multi-sensor nodes which have weak computing power and high real-time requirements.Therefore,it is significant to study the multi-channel EEG-based compression sensing technology.Firstly,aiming at the complex process of traditional emotion recognition technology and the long processing cycle that does not apply to the on-board environment,the implementation platform of the emotion recognition system applied in the vehicle environment is designed.1 This section includes: acquisition based on analog front amplifying ADS1299 completed and data acquisition,Real-time processing and classification of EEG signals based on digital signal processor DSP,CAN bus module based on TL2515 sending classification results to on-board computer.The whole process is completed by the DSP,using a single thread to complete,the process is simple,fast,and real-time.Secondly,the actual demand for wearable vehicle-borne brain EEG acquisition channels should not be excessive.By analyzing the data in the DEEP database of emotions and emotions,the characteristics of pleasure and unpleasant emotions are analyzed and it is found that the percentage power spectral densities of the two emotions in the ?,? and ? bands are significantly different.As a classification feature,the two emotions are clearly distinguishable from the classification accuracy.According to the analysis of brain energy of 32 collection pathways in the database,seven EEG acquisition pathways are identified,FP1,FT7,FT8,CP5,CP6,O1 and O2 in the frontal,parietal and occipital regions.According to the features extracted from different emotions,the two classification methods of support vector machines are used to train emotion classifiers in MATLAB.The classification accuracy rate is 71.6%.The FFT of the butterfly algorithm is used to extract the features in the DSP to improve the operation speed.The processing time for 1024 data points is 9.2 microseconds.Finally,a single individual is tested and tested for emotion induction,and compared with the characteristics of the DEAP database.The characteristics are adjusted according to individual differences and the classifier wis retrained.The classification accuracy rate is 65.2%.Simulated autopilot environment is used for performance verification,and 7 channels of EEG data are classified and judged in DSP.And the result of the 7 judgments as the final classification result,the classification accuracy rate was 67%.