Mental State Research Based On The Fusion Of Multi-source Physiological Information

With the development of brain-computer interface(BCI)technology and artificial intelligence(AI)algorithms,the study of human mental state based on EEG signals has become one of the hot research issues in the world.The mental state of the human body will have a direct impact on work efficiency,mental concentration and decision-making ability.In some special posts and important fields,mistakes caused by poor mental state will lead to irreversible and serious consequences.In order to avoid the occurrence of such errors,in recent years,multisource information fusion based on EEG signals supplemented by bioelectrical signals has become a hot research direction.This mode makes use of the complementarity of multi-source information and makes some achievements,but it still has the following problems: the collected signal contains a lot of noise interference which is not easy to remove,and the classification algorithm has low recognition rate.To solve these problems,a human mental state assessment system based on EEG and ECG signals is studied.1.Study on EEG and ECG data preprocessing and feature extraction.Chebyshev notch and median filter were used to remove power frequency interference and baseline drift respectively.Aiming at the EEG artifact in EEG signal,the VMD algorithm was used to conduct the denoising experiment on the semi-simulated EEG pollution model,and the feasibility of the proposed method was proved.Wavelet packet transform was used to reconstruct the preprocessed EEG rhythm waves,and the power spectrum of each rhythm band and the differential entropy representing the signal complexity were calculated as the EEG characteristics.In terms of ECG signal feature extraction,the RST detection algorithm based on state machine is used to locate R wave in QRS wave group,and other waveforms are located by their relative position with R wave.The mathematical statistics of each waveform,heart rate,heart rate variability and HRV high and low frequency are used as the characteristic indexes of ECG signal.2.The classification algorithm in the field of mental fatigue recognition is studied,and a DBN-GA-BPLM algorithm is proposed.In this algorithm,the deep belief network DBN is used to perform data fusion for the multi-source physiological features of the feature layer,and the high-dimensional feature vectors are mapped to the low-dimensional space so that they can be directly used as the input feature vectors of the classifier.Based on the traditional BP neural network,the classifier uses genetic algorithm to optimize its initial parameters and thresholds,so that the initial conditions of each training remain unchanged.In the aspect of neural network training speed,LM algorithm is used to speed up the learning process of the algorithm.Experimental results show that the training speed of this method is slow in the first training,but the accuracy of fatigue state recognition is 90.8%,which is better than most classifiers.3.The emotion classification algorithm is studied,and a DBN-SVM algorithm is proposed to recognize the three-classification emotion.In this algorithm,the deep belief network DBN is used to carry out data on the multi-source physiological features of the feature layer,and the second-level SVM is used as the classifier to classify emotions.The experimental results show that the accuracy of this method is 87.5%,and the classification effect is ideal.