Multi-frame decision fusion based on evidential association rule mining for target identification

In the multi-sensor target identification problem involving multiple frames, it is important to fuse the potential information characterizing inherent relations among frames with uncertain decision inputs for enhancing the decision-making process. However, due to the influence of environments or other interference factors, the priori knowledge that accurately represents these relations is usually hard to obtain. To overcome this difficulty, we propose a rule mining-based multi-frame decision fusion (abbreviated as RMDF) method, in which the unknown relations can be discovered from a series of historical sensor reports in the framework of belief functions. First, to accommodate data uncertainty, new measures of evidential support and confidence are defined for a constructed multi-frame evidential database, which are generalizations of the support and confidence measures in binary and probabilistic databases. Then, with these measures, an evidential association rule mining algorithm is developed to discover the relations among frames from a series of historical reports. Finally, how these mined rules are properly combined with uncertain decision information using belief function theory is explored. The key benefit of the RMDF method is that it enables modeling the uncertain relations among frames for deriving more accurate decision results. To demonstrate the feasibility and effectiveness of our proposal, an airborne target identification problem is studied under different conditions and the numerical results show that the identification performance of our method is significantly better than the traditional expert knowledge-based method where the available knowledge is inevitably incomplete or inaccurate.