Research Of Decision Fusion Algorithm Based On DS Evidence Theory

As an emerging subject applying in target tracking, pattern recognition, medical image processing and earth science,decision fusion can achieve a better decision-making performance through utilizing the redundancy and complementarity of a variety of different information obtained by multiple sensors. DS evidence theory is a typical decision fusion algorithm, which is known by many scholars due to its good measurement and reasonable description of system's uncertainty.This paper thoroughly studies the decision fusion algorithm based on DS evidence theory mainly by the following parts:Firstly, we expound the theoretical foundation of DS evidence theory, principles and properties of DS combination rule. Among them, the paradoxes that often appear in multi-sensor monitoring environment are analyzed emphatically. Furthermore, we attribute paradoxes to inaccurate acquisition method of evidence models and evidence conflict.Secondly, the acquisition method of evidence models was studied. We introduce 3 classic acquisition methods. Thinking about their low practicability with the need of prior knowledge,we introduce Deng's gray relation theory to raise a novel acquisition algorithm of evidence models based on slope correlativity to increase acquisition precision. On this basis, taking the different contributions of different features to evidence model into consideration, an innovative acquisition algorithm based on weighted coefficients is presented. The simulation results demonstrate, two proposed acquisition methods both get rational evidence models,which reflect sensor's support to different propositions well,and the supports among different propositions are apparently diverse, which is benefit for decision-making and system recognition. Besides, the latter has better performance.Thirdly, solution of evidence conflicts was studied. We discuss 12 common solution methods respectively, and choose the weighted evidence combination method which gets the best effect as comparison algorithm. After these analyses, we put forward a combination algorithm based on Minkowski distance and a combination algorithm based on discount processing by the introduction of Minkowski distance and discount factors separately. The simulation results indicate, two proposed algorithm can not only solve conflict reasonably and effectively, but also get fusion results beneficial to decision system. Compared to the weighted evidence combination method, they both enhance the validity and reliability of synthetic results. Besides, the latter has better performance.Finally, based on the study of paradoxes, we bring the decision fusion algorithm based on DS evidence theory, and apply it to emitter identification field. Ulteriorly, the diagram of this innovative decision fusion algorithm is given. This decision fusion method realizes signal identification by feature extraction and fusion of multi-period measurement and multi-dimensional of signals' entropy characteristics. The simulation results manifest, the decision fusion algorithm based on DS evidence theory can improve the emitter recognition accuracy and enhance the robustness of decision system. Thus, it verifies the superiority and practicability of this modified decision fusion algorithm.