To precisely estimate the micro-motion parameters and realize target identification under high order motion, we propose a separation and estimation method based on phase information. The main motion and micro-motion are separated by removing the polynomial phase signal part through the derivation of unwrapped echo phase. For the sinusoidal frequency modulation items with multi-dimensional micro-motion parameters after separation, an improved particle filter (PF) static parameter estimation method is proposed, and the efficiency of the algorithm is improved by designing the adaptive variance method and changing the number of particles. By designing cumulative residuals as a function of observed probability density, the simultaneous estimation of multi-dimensional parameters is realized in the nonlinear model. Simulation and experimental analysis verify the effectiveness and necessity of the proposed algorithm. The nonlinearity and the amount of calculation are reduced by processing the phase, which improves the anti-noise performance. The PF based parameter estimation method avoids the error transfer and improves the estimation accuracy effectively.