Abstract
Dempster–Shafer (D-S) evidence theory plays an important role in multisource data fusion. Due to the nature of the Dempster combination rule, there can be counterintuitive results when fusing highly conflicting evidence data. To date, conflict management in D-S evidence theory is still an open issue. Inspired by evidence modification considering internal indeterminacy and external support, a novel method for conflict data fusion is proposed based on an improved belief divergence, evidence distance, and belief entropy. First, an improved belief divergence measure is defined to characterize the discrepancy and conflict between bodies of evidence (BOEs). Second, evidence credibility is generated to describe the external support based on the complementary advantages of the improved belief divergence and evidence distance. Third, belief entropy is utilized to quantify the internal indeterminacy and further determine evidence weight. Lastly, the classical Dempster combination rule is applied to fuse the BOEs modified by their credibility degrees and weights. As the results of numerical examples and an application show, the proposed divergence measure can overcome the invalidity of the existing measures in some special cases. Additionally, the proposed fusion method recognizes the correct target with the highest belief value of 98.96%, which outperforms other related methods in conflict management. The proposed fusion method also displays better convergence, validity, and robustness.
Highlights
Multisource data fusion is widely used in the real-world applications, such as target recognition [1, 2], fault diagnosis [3, 4], and multicriteria decision-making [5, 6]
Inspired by pignistic probabilistic transformation in D-S evidence theory [55] and cross entropy in fuzzy set theory [6], we convert the basic belief assignments (BBAs) into probability distributions, which is in line with the classical Jessen–Shannon divergence [48], and we introduce the parameter t to ensure the validity of the logarithm value [68]
Target θ3 θ1 θ1 θ1 θ1 θ1 θ1 θ1 θ1 θ1 an improved method based on the Euclidean distance of BBAs; Deng et al [27] proposed a modified average method to combine belief function based on evidence distance; Yan et al [31] studied an improved belief entropy and its application in conflict management; Jiang et al [73] devised a combination rule based on Deng entropy; Li et al [38] proposed an evidence fusion method based on an improved Jousselme distance and Tsallis entropy; Yuan et al [37] designed a method combining Deng entropy and Jousselme distance to address conflict management; Xiao [1] proposed a multisensor data fusion based on the belief Jensen–Shannon (BJS) divergence and belief entropy
Summary
Multisource data fusion is widely used in the real-world applications, such as target recognition [1, 2], fault diagnosis [3, 4], and multicriteria decision-making [5, 6]. Because the characteristics of uncertainty, imprecision, and imperfection are inevitable in practical problems, determining how to model and handle such multisource and uncertain data is still an open issue [7,8,9]. To address this issue, many related theories and methods have been proposed and developed, such as probability theory [10], fuzzy set theory [11], rough set theory [12, 13], and evidence theory [14,15,16]. When bodies of evidence (BOEs) are highly conflicting, the Dempster combination rule, which is the core of D-S evidence theory, can produce counterintuitive results, such as the famous Zadeh paradox [19]. erefore, it is vital to effectively combine highly conflicting evidence in D-S evidence theory
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
