Correct situation awareness has a profound impact on pilot’s rational decision-making and the improvement in operation performance and flight safety. To effectively evaluate the state of pilot’s situation awareness, a deep learning method based on visual gaze information is proposed. The method extends the pilot visual gaze analysis channel and builds a deep neural network model for situation awareness state assessment. The model adopts a self-attention structure and takes visual fixation data as input to analyze the visual perception state of various flight elements such as altitude, heading and airspeed etc., and finally comprehensively evaluate and output the situation awareness state of pilots in real time. Experimental results show that the average accuracy of the model can reach 91.53%, which can significantly improve the accuracy of current pilot situation awareness state assessment. Meanwhile, the correlation analysis between model evaluation results and actual flight performance verifies the effectiveness of the model.