A New Improved D-S Evidence Theory Based on BJS Divergence in Multi-source Information Fusion

Multi-source information fusion is widely used in target recognition. Affected by interference factors, target recognition has great uncertainty. Relevant information cannot be accurately obtained only through a single evidence, which leads to unreliable results of target recognition. Dempster-Shafer (D-S) evidence theory is widely applied in multi-source information fusion due to the advantage of flexibility and effectiveness in combining plenty of uncertain information without prior probabilities. But if there are some high conflict elements in the evidence groups, it will produce some counterintuitive fusion effect, which is worse than before. To solve the above-mentioned problem, a new improved D-S evidence theory based on Belief Jensen-Shannon (BJS) divergence was proposed. We first utilize BJS divergence to locate the evidence groups of which discrepancy exceeds a certain threshold, and the threshold can be set according to the requirements of evidence groups. Subsequently, let the evidence group of which the BJS divergence exceeds the threshold change into the mean value of all the evidence events, this approach can evenly distribute the degree of conflict. Finally, the fusion results can be calculated by the combination rules of D-S evidence theory. Some experiments indicate that this method can get more reasonable results.