A belief Rényi divergence for multi-source information fusion and its application in pattern recognition

Multi-source information fusion technology has been widely used because it can maximize the use of information that collected from multiple data sources for decision fusion. As an uncertain information processing theory, Dempster-Shafer (D-S) evidence theory is prevalent in the field of multi-source information fusion. However, there is still room for improvement in the handling of uncertain problems involving highly conflicting evidence sources. For the purpose of improving the practicality and efficiency in handling highly conflicting evidence sources, a new belief Rényi divergence is defined to measure the discrepancy between evidences in D-S evidence theory. The proposed belief Rényi divergence takes belief function (Bel) and plausibility function (Pl) into account, thus allowing it to provide a more rational and telling approach for measuring differences between evidence. Moreover, some important properties of belief Rényi divergence have been studied, the belief Rényi divergence regards to Hellinger distance, Kullback-Leibler divergence and χ2 divergence, which ensures that the metric has a wider range of application scenarios. Based on the proposed belief Rényi divergence measure, a novel multi-source information fusion method is designed. The proposed belief Rényi divergence is used to measure difference between evidence; Deng entropy is used to quantify the uncertainty, thereby calculating information volume of the evidence. Accordingly, the proposed method can fully assess relationship among evidences and information volume of each evidence. Through a comprehensive analysis and experiments, practicality and effectiveness of the proposed method for multi-source information fusion are verified. Finally, an iris dataset-based experiment is implemented to verify the new proposed divergence measure and the multi-source information fusion algorithm has a more extensive applicability.