A particle filter method based on fruit fly optimization algorithm is proposed to alleviate the sample impoverishment caused by resampling. When fruit flies forage, they usually fly from low concentration areas to high concentration areas efficiently and constantly. This optimum process is introduced into the particle filter to drive particles towards the high likelihood areas ceaselessly, and thus improves the overall quality of the particle swarm. Considering that the premature convergence is always associated with the fruit fly optimization algorithm, crossover and mutation operations of genetic algorithms are applied herein adaptively to keep the diversity of samples. Firstly, the particle distribution is improved by cross operation. When the algorithm falls into the local optimum, the Cauchy mutation perturbation is then used to help the fruit fly optimization algorithm jump out of the local optimal point effectively and continue searching for global extremum. The nonlinear simulations and target tracking experiments show that the proposed algorithm improves the estimation accuracy of the nolinear systems state, and it has better stability and reduces the number of particles required for state estimation at the same time.