In recent years, the fully automated system is still a huge challenge although automation technology and information technology have been rapidly developed. Human operator is still widespread in all kinds of human-machine systems, and has become the key to the safe operation of the system. In safety critical situations, there has been a growing concern about the operator functional state (OFS) breakdown under excessive level of mental workload (MWL). High risk operating task in complex human-machine system is vulnerable to the operator’s breakdown of functional state. One of the potential solutions is to well classify the mental workload of operator.However, in the classifier, with the increase of the dimension of the input feature, it’s accuracy will be affected. In MWL classifier, high dimension of Electrophysiological signals made the problem very difficult, and the results of different classifier would be different for the same problem. In this paper, the MWL data from independently designed and completed experiments was processed by filtering with a3-order Butterworth filter; EOG removal with Coherence method; Fast Fourier Transform into frequency domain; and then dimension reduction through KPCA; at last, we classified the data with Kernel Fisher Discriminant Analysis (KFDA).The results show that the KPCA algorithm can effectively improve the classification accuracy, and the KFDA classifier has advantages over Support Vector Machine (SVM) in multi-class classification. |