Research On The Technology Of Data Fusion Based On Evidence Theory And Neural Network

With the rapid development of data fusion technology, the processing of uncertain information has been a research hotspot. Evidential reasoning, as one of the effective inference methods processing uncertain information, is widely applied to data fusion. In this paper, the theory of evidential reasoning is studied thoroughly; method of target identification is designed using combination of evidential reasoning and neural network technology.The main work and achievements are as follows:Firstly, the basic principles, the fusion process and structural model of multi-sensor data fusion are introduced, three fusion structure of target identification (properties) level of multi-sensor data fusion are discussed, and the data fusion algorithm is summarized.Secondly, the basic theory and the latest developments of evidential reasoning are introduced. The problems of evidential reasoning in application are given.Considering that evidential reasoning can not resolve problems that there are serious conflicts between evidences, many scholars have put forward a variety of methods, for example the method of Yager.In this article classification for typical algorithm is given. The classification method can effectively distinguish the basic similarities and differences between some evidetial reasoning approachs, provided further direction for evidential reasoning.Thirdly, the feasibility that neural network and technology of data fusion are integrated is discussed; the process of data fusion based on neural network is given.BP algorithm is emphased.Fourthly, the nearest neighbor method is used to compare effect of target identification, in which we use moment invariants before and after amendment.Finally, considering that it is difficult to obtain Basic Probability Assignment, two methods of forming Basic Belief Assignment fuction are proposed. The simulation result show that the method based on technology of neural network is better. It can get over the subjectivity of existing methods that depends on the experience of experts. A method of improving belief degree by use of evidence reasoning to combine multiple images arising from the same targe is studied. Recognition effect is compared by using various combination rules. The simulation result testifies the feasibility and validity of method. Research work of the dissertation is a bold and useful attempt for the development of multi-sensor data fusion, it is helpful for data fusion、neural network、target identification and so on, it has important academic meanings and project application values.