Concerning the low accuracy and convergence of standard Particle Swarm Optimization (PSO) algorithm, an improved particle swarm optimization based on particle filter re-sampling and mutation named RSPSO was proposed. By using the resampling characteristic of abandoning particles with low weights and duplicating and retaining particles with high weights, an existing method for mutation was adopted to overcome the disadvantage of particle degeneracy, which greatly enhanced the local search capability in the later searching stage of PSO algorithm. RSPSO algorithm was compared with the standard algorithm and some other improved algorithms in the literature under different benchmark functions. The experimental results show that RSPSO has faster convergence, higher accuracy and better stability, and it is able to solve multi-modal problems globally.