Pilots' fatigue status could influence aviation safety. The recognition of fatigue status of pilots is of utmost significance. We proposed a new deep learning model via analyzing electroencephalogram signals to reduce the complexity of feature extraction and improve the accuracy of recognition of fatigue status of pilots. We firstly applied wavelet packet transform to decompose electroencephalogram signals of pilots to extract the δ wave (1∼3 Hz), θ wave (4∼7 Hz), α wave (8∼13 Hz) and β wave (14∼30 Hz), and the combined representation of them were de-nosing EEG signals. Then we used proposed deep sparse auto-encoding network to reduce the complexity of de-nosing EEG signals and gained learning features. Lastly, we applied Softmax classifier on learning features and the experimental results showed that the proposed deep learning model had a nice recognition, and the accuracy of recognition was up to 91.67%, which meant that the proposed method performed excellently compared with the state-of-art methods.