Research On Several Algorithms Of Data Fusion

The basic idea of data fusion is to synthesize the data from different sources, modes, media and time by some specific rules, so as to obtain more accurate description of the object. It has gained popularity over past decades with the advent of vigorous sponsorship. In recent years, data fusion has attracted more and more attention and also been applied to many fields. Data fusion is such a kind of theories and methods that can process data from different sources. By synthetically treatment with these data, data fusion can produce much more exact presentation of real environment. Essentially, data fusion is just an issue of algorithm. So it is of great importance to research data fusion algorithm. This paper studied the results of domestic and foreign papers in detail. The main contributions in this dissertation can be summarized as follows:Firstly, this paper introduces the definition, system structures, and applications of data fusion. Secondly, through examples, two uncertainty reasoning theories of Bayes and D-S evidence are studied.Various improvement methods of D-S evidence theory are summarized. Thirdly, rough sets theory is introduced systematically, and is compared with fuzzy set theory and D-S evidence theory. And its theory and method is applied into an instance of data fusion.Forthly, the design and training of BP neural net are discussed intensively. The advantages and disadvantages of data fusion based on BP neural net are summed up by an instance.Lastly, considering the ability of rough sets theory for reduction of decision system and that of neural networks for clustering and nonlinear mapping, a hybrid system of rough sets and neural networks for data fusion is presented. Firstly, the continuous attributes of sample data are discretized with self-organizing map neural network.Then, reduction are performed based on rough sets theory, and the rules are extracted. Lastly, according to the reducted sample data, BP neural network is designed.The final result of the system is outputed by fusing the result of rough set data analysis and that of BP neural net.This method is simulated and the experimental results demonstrate its effectiveness.