基于相似度动态修正的冲突证据组合方法

Combination Method of Conflict Evidences Based on Similarity Dynamic Corrected

针对信息融合中冲突证据组合时易出现的一般冲突、一票否决和鲁棒性等常见问题,有两类改进策略:一类修改DS(Dempster-Shafer)组合规则,另一类修改证据源模型.提出一种基于封闭世界的修改模型方法.引入Jousselme距离函数来量化焦元属性及证据之间的相互关联性,进而计算各证据的支持度.对证据支持度进行加权平均后得到参考证据,利用该参考证据对各原始证据进行不确定性判定,获得各原始证据与参考证据之间的大小相似度和方向相似度.在此基础上建立一个相似度动态修正模型,利用DS组合规则进行证据组合,对动态修正模型的多组组合结果求平均作为最终结果.通过仿真实验验证所提出方法的有效性和合理性.

Two traditional improvement strategies,modified DS（Dempster-Shafer） combination rules and modified evidence model,are used to solve some typical combination problems of conflict evidences in information fusion subject,such as general conflict,vote-down,and robustness.A new method of modified model based on closed world assumption named CESDC（Conflict Evidence Similarity Dynamic Corrected） is proposed in this paper.Jousselme distance function is introduced to quantify the correlations between focal attributes and evidences,based on which the support degree value of each evidence could also be calculated.Referenced evidence mentioned in this paper is obtained by the weighted average of all evidences according to their support degree values.The referenced evidence is used for uncertainty verification to count out the value similarity and direction similarity between each evidence and referenced evidence.It also establishes a dynamic correction model of the value similarity and direction similarity,the corrected result of which uses the DS combination rules to achieve the evidences combination.Then we set the average value of those multi-group results,which are all computed by the dynamic correction model,as the final fusion result.Experiment results show the effective and reasonable of the proposed method.