一种基于“交并集”和Pignistic概率的证据冲突的改进方法

A New Solution Based on “Conjunctive & Disjunctive Pooling” and Pignistic Probability Transforms According to the Evidence Conflict Problems in D-S Theory of Evidence

针对D-S证据理论无法解决高冲突证据的缺陷,通过对现有几种证据冲突的改进方法进行分析,提出了基于“交并集”和Pignistic概率的改进方法。本文放宽D-S组合规则的假设,只要求证据在组合时至少有一条是真实的,如果证据A和B相互支持,说明它们都是真实的,可以用“交集”运算将证据的信度聚焦在它们的交集上;如果证据A和B相互冲突,表明不知道哪一条证据是真实的,则用“并集”运算将信度聚焦在它们的并集上,即证据支持A或B中的一个,这种思路更符合人类的直觉。由于在目标识别系统中,最终决策是单个待识目标,因此以还要用Pignistic概率转换法将多元素命题的BPA再分配给它的各个组合元素,最后,信度最高的元素作为结果进行输出。实验表明,本文方法在解决证据冲突方面较其他方法拥有明显的优势。使用本文方法时,证据的融合顺序对融合结果没有影响,因此可以很方便地编程实现。

According to the defect that in the D-S Theory of Evidence, the evidence combination rules can＇t work correctly facing high conflicting evidences, a new solution based on ＂conjunctive ＆ disjunctive pooling＂ and Pignistic probability transforms is introduced. The solution supposes that at least one evidence is true among all the given evidences. When evidence A and B are consistent which means both the evidences are true, the beliefs of evidences will focus on their conjunctive pooling. On the other hand, when evidence A and B are inconsistent which means can＇t judgewhich evidence is true, the beliefs of evidences will focus on their disjunctive pooling. In object recognition systems, owing to the request of single final output, a pignistic probability transform is used to reassign the basic probability assignments of multi-element propositions to each element thus the final output is the one with the highest belief. The experiment results show that the solution can get best performance evaluation. Finally, the sequence of evidence fusion has no effect on fusion results so the solution can be programmed easily.