Search

Since Dempster–Shafer evidence theory was proposed, it has been widely and successfully used in many fields including risk analysis, fault diagnosis, wireless sensor networks, health prognosis, image processing, and target tracking, etc. However, many counter-intuitive results of data fusion will be obtained when evidence fused is highly conflicting. So far, this is still an open issue. To address this issue, many methods have been proposed, but they have not been comprehensively summarized in recent years. In this paper, a detailed survey is set forth about the optimization and application of evidence fusion algorithms based on Dempster–Shafer theory. Firstly, the principle of Dempster–Shafer evidence theory is introduced comprehensively. Then, the existing researches on modifying combination rule and pre-processed pieces of evidence to solve the counter-intuitive problem are reviewed in detail. Next, the performance of these studies is demonstrated, deeply analyzed, and discussed through experiments on general examples. And finally, the application of Dempster–Shafer evidence theory in different fields is critically summarized. What is more, analysis of the current status and the development trend of the research on evidence theory are concluded, which can provide a more comprehensive understanding of the development of the Dempster–Shafer evidence theory. • This survey is comprehensive about the methods based on optimization of D–S theory. • This paper has certain necessities, aiming at the latest research of evidence theory. • Trials are included to provide an objective comparison among typical methods. • This work introduces theories, applications and the development trend of D–S theory. [ABSTRACT FROM AUTHOR]

Dempster–Shafer evidence theory can combine multiple sources of uncertain information to form a consistent basic probability assignment (BPA). How to transform BPA into decision probability is still an open issue. To solve this problem, an importance-based Decision Probability Transformation (DPT) method is proposed in this paper. In this method, the importance weight is introduced into the belief interval for the first time, and then the importance is used to define the information volume and superiority degree, so that more reliable decisions can be achieved through the trade-off between conservative and aggressive. Comparative experiments on numerical examples and applications show that the proposed DPT method can weigh the aggressiveness of decisions, and its decision probability transformation behavior is the most rational and its application performance is the most reliable. Especially, the accuracy rate of classification application is 95.31%, which is at least 0.54% higher than other compared methods. • Importance is introduced for the first time to weigh the aggressive of decision. • The information volume and superiority degree are defined with importance weights. • Weighing the importance of beliefs yields the most reliable decision probabilities. • The proposed method performs best in target recognition and classification. [ABSTRACT FROM AUTHOR]