The Research On Discretization Of Continuous Variables And Weight Assignment In Multi-source Information Fusion

With the rapid development of multi-sensor sensing technology,intelligent computing technology and wireless communication technology,the military pioneered the concept of data fusion,which refers to the process of integration of multiple data and knowledge representing the same real-world object into a consistent,accurate,and useful representation.The goal of data fusion is to combine relevant information from two or more sources into a single one that provides a more accurate description than any of the individual data sources.In recent years,multi-sensor data fusion has received significant attention for both military and nonmilitary applications,so the research on multi-source information fusion(MSIF)technology has been widely concerned.The research contents of this thesis are summarized as follows:1.Research on the discretization of continuous variables(DCV)in MSIFIn this thesis,we focus on the Bayesian network based on probability theory.In order to solve the problem that the current algorithm for DCV is not accurate enough,this thesis proposes a new method of DCV based on the probability assignment for the standard states with the equal expectation,which is more reasonable and scientific to achieve the processing of the continuous variable without any data loss.This method provides a more accurate and effective way to realize DCV.2.Research on the multi-variable weight assignment(MVWA)in MSIFMulti-variable information fusion(MVIF)is the main content in MSIF.Generally,different variables have different influence to the decision-making.Most of the existing methods of weight assignment relay too much on expert knowledge or pay excessive attention to statistical characteristics.This qualitative analysis lacks theoretical basis and can't meet the requirements of the modern data fusion.On the basis of information theory,this thesis proposes a new method of MVWA based on information gain,which can quantitatively conduct weight assignment by information gain.3.Research on weighted Bayesian network algorithmBayesian network algorithm is based on probability theory,and its parameter learning is the supervised.In order to gain the more accurate reasoning results,this thesis proposes two methods of introducing weight into Bayesian network.One is conducted at the feature level,which adds weight information into the evidence for the network reasoning.The other is conducted at the decision level,which adds weight information into the reasoning result of each variable because of the support of Bayesian network for incomplete evidence.This algorithm can effectively improve the inference accuracy of weighted Bayesian network algorithm.By the research on DCV based on the probability assignment for the standard states with the equal expectation,the proposed algorithm can achieve the processing of the continuous variable without data loss and apply the Bayesian network reasoning into the processing of continuous variables in MSIF.By the research on MVWA based on information gain,the proposed algorithm brings the theoretical basis into weight assignment and can achieve the quantitative weight assignment of continuous variables.By the research on weighted Bayesian network algorithm,the proposed algorithm can effectively improve the accuracy of Bayesian network inference and provide a solid foundation for the popularization of Bayesian network.