The physiological signals collected in different states during the experiment are filtered and denoised, and the initial data are preprocessed, so as to improve the jumble of data. And calculating and analyzing the characteristic physiological parameters of EMG signals, heart rate and heart rate variability of ECG signals and other time-frequency domain indexes; Analyze the phase average, standard deviation, maximum value and other time-frequency domain indexes of skin electrical signals and respiratory signals. Organize the multimodal physiological parameters of the participants, which is convenient for later machine learning analysis. Integrating and counting a plurality of physiological characteristic parameters, calculating Kendall correlation coefficient, and removing parameters with excessive correlation; PCA algorithm is used to fuse features and reduce dimensions of multi-physiological signals to increase the accuracy of model analysis. In the flight fatigue monitoring, the machine learning method based on LVQ is used to identify the fatigue of participants. In order to further improve the recognition accuracy and reliability of the system, the same data set is input into SVM model and BPNN model for fatigue recognition. The flight fatigue detection results and model prediction accuracy of three different classification modes are calculated and compared, and the best model is selected.