The pilots'cognitive states are essential factors affecting their control performance to the flight control system. Generally, cognitive states can not be measured directly. However, they can be gained indirectly by the means of physiological signals. Based on the characteristics of the physiological signals in time frequency domain, wavelet analysis is used for establishing the feature sets, and an FPA optimized Gaussian Process (GP) model is proposed for classification based on flower pollination algorithm, which is used for analyzing the pilots'cognitive states in a full flight simulation. Through the comparison between the classification results and the NASA-TLX test result of the pilots, the validity of this method is verified.