With the rapid development of civil aviation industry in the world, aircraft has become the first choice for people to travel in daily or long distance. The development of civil transport aircraft provides convenience and speed for people’s work and life. But at the same time, the incidence of aircraft safety accidents has been high. Through the investigation of the causes of the accidents, it is found that the aviation safety accidents caused by pilot fatigue account for a high proportion in the total number of accidents. Therefore, it is of great practical significance to scientifically and efficiently monitor the physiological and psychological fatigue of pilots in flight to reduce the incidence of aircraft safety accidents. Based on the current research theories and foundations at home and abroad, the most mainstream and representative method is to monitor the real-time physiological signals of pilots in flight. In this paper, the EMG, skin electricity, ECG and respiration signals of pilots during a long simulated flight are collected through experiments. After filtering, preprocessing and calculating the collected data, the time-frequency domain characteristics of four different types of physiological signals are obtained, totaling 29 characteristic indexes. Multi-modal feature screening is used to reduce the dimension of the data, and machine learning technology is used to find the most accurate model suitable for pilot fatigue monitoring, which can effectively identify the fatigue state of pilots, so as to achieve timely and accurate monitoring and evaluation of pilots’ psychological and physiological fatigue. The research in this paper provides a new method and feasibility for the multi-modal fatigue state monitoring of pilots.