深度稀疏自编码网络识别飞行员疲劳状态
Using deep sparse auto-encoding network to identify pilots’ fatigue status

针对飞行员疲劳状态识别的复杂性和准确性，提出一种基于脑电信号的深度学习模型。首先对飞行员脑电信号进行滤波分解，提取delta波（0.5~4Hz）、theta波（5~8Hz）、alpha波（7~14Hz）、beta波（14~30Hz），提取基于脑电节律波的频域特征，作为识别模型的输入向量。其次，将一种基于深度稀疏自编码网络-Softmax模型用于飞行员疲劳状态识别，并与单层的稀疏自编码网络-softmax和传统方法PCA-Softmax模型识别结果进行比较。最后，实验结果显示，针对飞行员疲劳状态识别问题，所建立的学习模型具有很好的分类识别效果，具有较好的工程推广价值。
Aiming at the complexity and accuracy of recognition of pilot’s fatigue states, a deep learning model based on electroencephalogram signals was proposed. Firstly, EEG signals were filtered and decomposed, and the delta wave (0.5 ~ 4Hz), theta wave (5 ~ 8Hz) ，alpha wave (7 ~ 14Hz) and beta wave (14 ~ 30Hz) were extracted and the frequency domain features of four rhythms were also extracted as the input vectors. Secondly, a deep sparse auto-encoding network-Softmax model was proposed to recognize pilots’ fatigue states. Its recognition results were compared with those of single layer sparse auto-encoding network-softmax and traditional PCA-Softmax model. Finally, the experimental results showed that the learning model has a good classification for the recognition of pilots'' fatigue states, and has a good value for engineering promotion