| With the rapid development of modern science and technology,the mental work of operators in man-machine system is gradually increasing,and the long-term mental consumption will lead to higher mental workload.Studies have shown that high mental workload will accelerate operator fatigue,which will lead to human safety accidents and adverse social impact,while low mental workload will lead to the waste of resources and unstable performance.In addition,because EEG signals can reflect the changes of mental workload accurately,it is necessary to extract and screen the features of EEG signals and classify the mental workload effectively.The main work of this paper is as follows:(1)Aiming at the problem that most mental workload classification methods simply use the four rhythms((?)? rhythms)of EEG signals,this paper uses Gini impurity for feature selection,and the highest sensitivity under different mental workloads was found to be ? rhythm,and the mental workload classification experiment was carried out in combination with the time domain features.The experimental results show that the use of the energy features of rhythms and variance can effectively classify the mental workload,which proves the feasibility of the method and reduces the feature redundancy.(2)Aiming at the problem that there are too many electrode channels on the EEG acquisition equipment,and most channels have information redundancy and noise interference,which reduces the efficiency of the classification model,an EEG channel selection method based on DB index is proposed.The EEG characteristics of each electrode channel are calculated and sorted by the DB index,and the results obtained by all data are counted to obtain the optimal channel set,which reduces the noise and irrelevant information caused by invalid channels.By comparing the results of using the optimal channel set and using all channels for mental workload classification,it is found that the optimal channel set can achieve better classification accuracy,which demonstrates the effectiveness of the proposed EEG channel selection strategy in this study.(3)Since the EEG signal is a mixed signal recorded from multi-channel electrodes,in order to mine the relationship between the EEG signal and the mental workload,and extract the characteristics of the signal more accurately,this paper uses the Fast ICA algorithm to obtain pure EEG independent components,and the isolated EEG independent components were used for feature extraction and mental workload classification.By comparing the results of the mental workload classification experiments,it is found that the classification accuracy obtained by extracting the features of the EEG independent components is better than the results obtained by extracting the features of the mixed EEG signals,which demonstrated the accuracy and reliability of mental workload classification using EEG independent components.(4)A mental workload classification model based on SVM is established,and a binary classification model is established to distinguish low workload and high workload states by using multi-dimensional features.In order to select the appropriate kernel function,the linear kernel function and the Gaussian kernel function were used to construct the classification model,and the grid search was used to select the best combination of parameters,and the K-fold cross-validation method was introduced to suppress the overfitting of the model.(5)Aiming at the problem that the data across time does not necessarily have the same distribution,and it is necessary to collect new data to rebuild a new classification model,this paper proposes to use the TCA algorithm to reduce the distribution difference between the data.By comparing the classification accuracy of direct mental workload classification and classification using TCA,it can be seen that the TCA algorithm can reduce the distance of data distribution between EEG data obtained in different periods and improve the capability of the classification model. |
PDF Full Text Request