An Improved Evidence Classification Synthesis Method Combined Information Entropy

In the current era, the multi-sensor data fusion is widely used in more and more fields. Due to a variety of factors, the science and technology are progressed and developed and the large amounts of corresponding data fusion algorithms arise at the right moment, such as BP neural network algorithm, evidence theory algorithm,Bayes theory algorithm and so on. As a kind of quite classical data fusion algorithm which deals with uncertainty issues, evidence theory is developed for forty years and is further stated by all levels of important domestic and foreign scholars.This article concentrates on D-S evidence theory and combination rule, and carries on the thorough analysis on them. The traditional evidence theory may lead to the result conflicting with the evidence in dealing with high conflict evidence, which results in the failure of target decisions. A plenty of existing literature take study and improvement on this deficiency. This paper in line with the deficiency of the solutions about the conflict evidence issue proposes an improved solution.On the deficiency that D-S evidence combination rule deals with the conflict between evidences, this paper in the basis of the classification between evidences classifies and correct the similarity attribute between evidences and the information entropy for mass evidences. The similarity attribute between evidences can be divided into three seed properties, respectively the distance between evidences, the conflict amount and the angle. The uncertain attribute of the evidence is described by the information entropy attribute.Combining the similarity attribute between evidences and the information entropy attribute of evidence itself, the evidence set could be afresh divided into high credibility evidence, general evidence and conflict evidence. The sorted evidence set is given different importance coefficients, and is modified to improve. After modifying, the general evidence and high conflict evidence are closed up to the high credibility of evidence opinions. Then the revised evidence is synthesized.Incorporating with four target attributes in greenhouses(respectively the air temperature, soil temperature, air humidity and co2 concentration), the cucumber growth environment is evaluated by experiment using improved fusion algorithm in this paper. The corresponding basic probability distribution function of each sensor iscomputed by rough set theory. The experimental results that the proposed compared with other improved methods show that the improved method can not only effectively solve the problem of conflict, but also reduce the uncertainty of evidence. This approach can do better data fusion. And the feasibility of this innovative method and the rationality of the model are determined.