An improved multisource data fusion method based on a novel divergence measure of belief function

How to manage conflict in Dempster-Shafer (D-S) evidence theory is still an open problem. To address this problem, a novel divergence measure is proposed to measure the distance between evidence. The proposed divergence measure comprehensively considers the difference between sets of belief function and creatively deal with possible zero in the denominator by pre-averaging with base belief function. It satisfies symmetry, nonnegativeness and nondegeneracy. Furthermore, some numerical examples demonstrate that the proposed divergence measure is more reasonable and effective compared with existing belief divergence measures. In addition, based on this proposed divergence measure, a novel fusion method for multisource data is introduced which considers both the uncertainty of the evidence itself and mutual support from other evidence. The proposed fusion method achieves the highest accuracy compared with other existing fusion methods in the experiment and their time complexity is investigated in detail to distinguish them from each other. Finally, the proposed fusion method is applied to a real classification application and gains the highest accuracy in all three categories. Considering its high fusion accuracy and time cost, it is suitable for cases where accuracy is extremely crucial and immediacy is not strictly required. Therefore, it is an effective multisource fusion method on realistic complex cases.