摘要目的:基于空间定向飞行模拟状态下飞行员的视觉注视行为，初步提出一种实时评估飞行员情景意识能力的方法。方法:通过抽签法随机抽取符合标准的歼击机飞行员。采用眼动仪收集飞行模拟器中飞行员的眼动踪迹特征数据，利用轻量级的YOLOv8n模型对训练中的感兴趣区域（area of interest，AOI）进行检测，构建AOI注视序列特征数据。记录飞行员的错觉体验和复杂状况改出结果，通过情景意识综合评估方法打分得到3个情景意识划分等级——优秀、良好、一般，作为标签数据。开发一种转换器与启发式模块融合的态势感知（transformer and inception module fusion situation awareness，Ti-SA）模型，用于提取和学习眼动踪迹时序数据与AOI注视时序数据的特征，并与其他多维时序分类领域的常用模型进行比较。结果:共纳入歼击机飞行员30名，均为男性，年龄23~38岁，飞行时间300~2 200 h。通过眼动仪共获取飞行员眼动踪迹的19个时序特征，飞行员的情景意识划分等级：优秀等级的12名，良好等级的15名，一般等级的3名。Ti-SA模型用于所得数据集，准确率为92.18%，精确率为92.95%，召回率为95.49%，F1分数为94.20%，优于其他多维时序分类领域的常用模型。结论:研究表明，所提出的数据集构造方法和Ti-SA模型能够有效评估飞行员在空间定向飞行模拟状态下的情景意识水平。

abstractsObjective:To propose a preliminary method for real-time assessment of pilot situational awareness based on assessing pilot′s visual gaze behavior during spatial orientation flight simulation.Methods:Fighter pilots who met the criteria were randomly selected by drawing lots. An eye-tracker was used to collect eye-track feature data from pilots in a flight simulator. The lightweight YOLOv8n model was used to detect the area of interest (AOI) in the training to construct the AOI gaze sequence feature data. The pilot′s illusory experiences and recovery from complex situations were recorded, and those were scored by the situation awareness global assessment technique to obtain such 3 situational awareness assessment levels as excellent, good, and fair which were used as labeled data. A transformer and inception module fusion situation awareness (Ti-SA) model was developed to extract and learn the features of eye-tracking time-series data and AOI gaze time-series data and was compared with other commonly used models in the field of multidimensional time-series classification.Results:Thirty fighter pilots were enrolled, all male, aged 23-38 years old, with flying hours of 300-2 200 h, were included in the study. Nineteen temporal features of pilots′ eye movement trajectories were obtained by eye-tracker. By situation awareness global assessment, 12 pilots were scored to excellent level, 15 to good level and 3 to fair level. When Ti-SA model was applied to the experimental dataset, the accuracy was 92.18%, the precision was 92.95%, the recall was 95.49%, and the F1 score was 94.20%, which were better than other commonly used models in the field of multidimensional time-series classification.Conclusions:The study indicates that the proposed dataset construction method and Ti-SA model can effectively assess the level of pilot situational awareness in spatial orientation flight simulation.