【摘要】 为解决接触式生理数据采集带来的低舒适性和对飞行员的干扰，提出1种基于非接触式面部识别方法和NASA工作负荷指数(NASA-TLX)的认知负荷评估方法来确保飞行员安全，对21名飞行学员在不同飞行阶段的认知负荷进行评估。首先，使用NASA-TLX量表来主观衡量飞行学员在不同飞行阶段所经历的认知负荷。随后，使用FaceReader提取面部表情特征进行特征分析，并使用分类模型对认知负荷水平分类。研究结果表明：LSTM,LSTM+Transformer和CNN+Transformer模型是认知负荷分类的有效工具，分类准确率分别为93.68%,94.64%和94.77%;随机森林特征选择和主成分分析在大幅降低特征维度的同时，可保持一定的分类准确性。研究结果验证了非接触式感知在识别飞行学员认知负荷方面的可行性，可为真实条件下飞行员认知负荷评估提供参考。更多还原

【Abstract】 To address the discomfort and interference on pilots caused by the contact-based physiological data collection, a cognitive load assessment method based on the non-contact facial recognition and the NASA Task Load Index(NASA-TLX) was proposed to ensure the pilot safety, and the cognitive loads of 21 flight students during different flight stages were evaluated.Firstly, the NASA-TLX scale was used to subjectively measure the cognitive loads experienced by the flight students during different flight stages.Subsequently, the facial expression features were extracted using FaceReader, followed by feature analysis, and the classification models were employed to categorize the cognitive load levels.The results show that the LSTM,LSTM + Transformer, and CNN + Transformer models are the effective tools for cognitive load classification, achieving the classification accuracies of 93.68%,94.64%,and 94.77%,respectively.The random forest feature selection and principal component analysis significantly reduce the feature dimensionality while maintaining a certain level of classification accuracy.The research results demonstrate the feasibility of non-contact perception in identifying the cognitive load of flight students, and can provide reference for the cognitive load assessment of pilots under real-world conditions.更多还原