Inference algorithm of variable structure DDBNs and multi-target recognition

The current inference algorithm on variable structure discrete dynamic Bayesian networks (DDBNs) suffers from the drawback of exponential growth in complexity as time slices increase. To solve this problem, this article introduces the basic idea of a forward-backward algorithm, and proposes a new inference algorithm of variable structure DDBNs. On the basis of an analysis of the data structure of the variable structure networks, the forward operator and backward operator of variable structure DDBNs are defined and the algorithm deduced in theory, whose amount of calculation has a linear relationship with the number of the time slices. In addition, the algorithm is applied to variable structure DDBNs to identify air multi-targets. By fusing the information of the "engagement behavior" node efficiently, the robustness of the identifying system is strengthened significantly. The validity of this algorithm is proved by the simulation results.