An atmospheric slices time domain signal generation algorithm model (feuilleté model) is proposed to simulate the signal under non-stationary atmospheric conditions. The wind inversion by means of timefrequency analysis is introduced, and the algorithm processing of signals simulated under non-stationary wind model is studied. Based on the actual system parameters, the echo signals are simulated under conditions of linear wind shear and national aeronautic and space administration (NASA) typical wind shear model, respectively. Wind inversion by analyzing the distribution of peak energy in the time-frequency domain after the application of Wigner-Ville transformation on the simulated signal is got. And in this process, the fluctuation of wind inversion, which results from speckle noise, cross terms and other factors, can be reduced by accumulating multiple time-frequency transforms of pulses. The results show that, wind inversion is obtained quickly and directly by utilizing time-frequency analysis methods and without division of range gate in detection zone. The fluctuation of wind inversion decreases after the processing of accumulation and average, and the similarity between wind inversion and input wind speed is improved significantly.