Target classification algorithm based on relevance vector machine and Particle filtering

Relevance vector machine (RVM) is one kind of intelligent classification algorithm with good performance. However, it is difficult to classify targets which have similar characteristics using RVM. In this paper a sort of dynamic classification algorithm which combines RVM and Particle filtering (PF) is proposed: When the characteristics of targets are resembling at initial state, this algorithm can accomplish the classification through estimating the states and trends from continuous observations at different time points; when the characteristics of targets are not similar at primary observation, RVM will be applied directly for target classification. Simulation results show that RVM-PF achieves a high correct rate with a few observations in the two situations above.