Research And Application Of Information Fusion Based On D-S Evidence Theory

The increasingly extensive application of wireless sensor network and the emergence of cyber physical systems make the information technology attract more and more attention.With the increasing complexity of sensor data sources,exploring a way to acquire information fast,comprehensively and accurately has become a hot topic.Therefore,the research on information fusion approach is of great significance.Considering the evidence theory has a strong reasoning ability under uncertainties,and can be used for multisource information fusion,the paper studies information fusion method based on evidence theory,presents not only a new method to measure conflict between evidences,but an improved approach to conflict evidence combination,and employs the latter in multisource information fusion.The specific contents are as follows.Firstly,the analysis and research of conflict measure.Dempster's combination rule,the kernel of D-S theory,is a classical rule to combine evidences represented by basic probability assignment function.However,when evidences highly conflict with each other,this combination rule will become invalid and obtain a counter-intuitive conclusion.In order to come up with a more effective method to conflict evidence combination,it is necessary to measure the conflict between evidences reasonably at first.Based on the summary of existing conflict measurement methods,the paper proposes a comprehensive method to measure conflict.On the one hand,the method considers the impact incompatibility between focals makes on conflict through direct conflict and potential conflict.On the other hand,the method takes into account the effect difference between evidences makes on conflict through the distance between betting commitment beliefs,thus to measure conflict comprehensively.Compared with existing conflict measurement methods,the method appears to be more effective,and provides a solid foundation for the further study on conflict evidence combination.Secondly,the study on conflict evidence combination.Currently,evidences produced by sensors are regarded to be reliable equally in most evidence combination methods,which makes evidence from low reliable sensors lead to the performance reduction in fusion system.To solve the problem,the paper introduces reliability index for evidence to describe certainty degree,and determine the discount rate foreach evidence according to the index,thus to reduce the impact of low reliable evidence on fusion result.In addition,conflict factor between evidences is obtained according to the proposed conflict measurement method,and consistent coefficient is acquired from membership function in fuzzy mathematics.By integrating the degree of conflict and consistency,the correlation degree between evidences is confirmed.Based on this,the weight coefficient for each evidence is determined.Then,discount rate and weight coefficient are used to revise evidence source,and an improved fusion approach is proposed.In the experiment,the improved method is used to identify the vehicle image information observed by sensors,comparative analysis with other methods shows the method can effectively identify the target,and has higher fusion accuracy,fast convergence speed.Finally,the research on information utilization issues in curve fitting for lifetime data in reliability data processing.By mining the feature information in depth in sample data,the improved evidence combination method is used to fuse the extracted information.According to fusion result,the sample points involved in fitting is obtained to fit the desired curve,and a valid reliability conclusion can be drawn.