Algorithm Of Information Fusion Based On Dependent Evidence

The theory of evidence is a kind of information fusion reasoning, which has perfect performance in solving the problem of uncertainty. With the theory of evidence being widely use in information fusion, conflicting evidence, correlative evidence and the generalizing of the theory of evidence are focus of application study of the theory of evidence .To solve the key problems of the theory of evidence, the main contributions are the follows:(l)The technology of information fusion and its fusion algorithm are introduced. The function and position of the evidence theory in the information fusion are explicated;the research advance of present evidence theory is systematically introduced, and current question and solving method are anal sized;(2)Conflicting and correlative evidence are studied which are often in application. Several classical improved-methods are analyzed and compared. With a new conflicting factor being introduced, the robust range is analyzed. It is proposed that the key of dealing with conflicting evidence is conflicting distribution;(3)The characteristic of the generalizing of the theory of evidence is discussed. The CDS and generalized evidence theory are introduced.(4)Research object with normal database, several methods of basic probability assignment function have been structured and applied multiclassificator. Combining neural network with correlative evidence, an integrated neural network system based on correlative evidence was built up and classification was done on the normal database. Valuable research results was provided for the application of the fusion of multiclassificators based on D-S evidence theory.Finally, all algorithms above are implemented in MATLAB, compare withthe test result given by improved algorithm and the result given by directly usage of D-S combination, indicate that the improving algorithms can utilize the advantage of neural network and correlative evidence, and better classification can accomplish by some special classification question, and indicate that the algorithms is effective.