Signal collected in flutter test with variable progression speed is usually non-stationary, both its frequency and amplitude changed over time, especially in the sub-critical state. The common non-stationary signal processing method, such as time-varying parameter modeling can hardly analyze nor track the mode of signal precisely under high non-stationary degree. Therefore, an adaptive particle filter method based on non-stationary degree is proposed. The tracking performance under high non-stationary degree of this method is verified by Simulation experiment data. The results suggest that method proposed in this thesis has better precision under high non-stationary degree when compared with original particle filter.