A reinforced final belief divergence for mass functions and its application in target recognition

As an extension of Bayesian probability theory, the Dempster-Shafer (D-S) evidence theory uses mass function instead of traditional probability distribution. This theory is famous for multi-sensor data fusion and can well represent uncertainty. However, if there are conflicting mass functions, the D-S evidence theory will fail. The existing methods for handling conflicting mass functions do not fully consider the interaction between focal elements. Therefore, to solve the conflict problem, this paper defines the similarity factor and quantity factor of the focal element and then considers the impact of their interaction. After that, we propose a novel reinforced final belief divergence (RFBD) measure to solve the conflicting problem in mass functions from the perspective of divergence measurement. We use several numerical examples to verify the superiority of RFBD in handling conflicting evidence under uncertain conditions. Finally, we combine belief entropy and ambiguity measure to propose the RFBD-based multi-sensor data fusion approach, then achieve target recognition in UCI datasets. The experimental results show that our RFBD is better than the advanced divergence methods currently available.