Evidence modeling and fusion method of conflict sensor information based on complex network

基于复杂网络的冲突传感器信息证据建模与融合方法

¹ School of Integrated Circuits and Microelectronics, Northwestern Polytechnical University, Xi'an 710072, China
² Shenzhen Research Institute of Northwestern Polytechnical University, Shenzhen 518057, China
³ Chongqing Innovation Center, Northwestern Polytechnical University, Chongqing 401135, China

Received: 16 June 2025

Abstract

Evidence modeling and fusion accuracy of multi-sensor information directly determines the target recognition performance. As a classic framework for uncertain information reasoning and fusion, Dempster-Shafer evidence theory has been widely applied in multi-source information fusion. However, when there is a strong conflict between evidence bodies, the direct application of Dempster's combination rule often leads to counter-intuitive and even unreliable fusion results. Although existing improved methods alleviate the conflict problem to a certain extent, they still have limitations such as slow convergence speed, insufficient ability to suppress interference from unreliable information, and redundant network modeling. To address the above issues, this paper proposes an evidence modeling and fusion method based on complex networks. The method maps evidence bodies to network nodes and introduces a dual-weight complementary modeling mechanism of direct and indirect weights based on the interrelationships between evidence. Specifically, the direct weights between network nodes are modeled by evidence distance to represent the similarity between evidence bodies, and the indirect weights reflect the indirect support relationships of evidence bodies in the network structure through cosine similarity. By fusing and normalizing the two types of weights, the adaptive correction of the original evidence bodies is achieved, and then Dempster's combination rule is used to complete the fusion of conflicting uncertain information. Experimental results show that the proposed method exhibits faster convergence speed of target evidence, stronger interference suppression capability, and more effective high-conflict evidence resolution performance in the multi-evidence fusion process, demonstrating good stability and reliability.

摘要

多传感器信息的证据建模与融合精度直接决定目标识别性能。Dempster-Shafer证据理论作为一种经典的不确定信息推理与融合框架, 在多源信息融合中得到了广泛应用。然而, 当证据体之间存在高冲突时, 直接采用Dempster组合规则往往会导致反直觉甚至不可靠的融合结果。现有改进方法虽然在一定程度上缓解了冲突问题, 但仍存在收敛速度慢、对不可靠信息干扰抑制能力不足、网络建模冗余等局限。针对上述问题, 提出一种基于复杂网络的证据建模与融合方法, 将证据体映射为网络节点, 并从证据间的相互关联出发, 引入直接权重与间接权重的双权重互补建模机制。其中, 网络节点间直接权重由证据距离建模为证据体之间的相似性, 间接权重由余弦相似度反映证据体在网络结构中的间接支持关系。通过对2类权重进行融合并归一化, 实现对原始证据体的自适应修正, 再采用Dempster组合规则完成冲突不确定信息融合。实验结果表明, 文中方法在多证据融合过程中表现出更快的目标证据收敛速度、更强的干扰抑制能力及更加有效的高冲突证据消解性能, 具有良好的稳定性和可靠性。