Target Recognition Algorithm And Its Application Based On Multi-source Information Fusion

Dempster-Shafer(D-S)Evidence Theory has the ability to express uncertainties and to fuse imperfect information.However,it has some limitations in dealing with high conflict information,which greatly limits its practical application.Therefore,the conflict measure algorithm and evidence fusion algorithm on multiple pieces of evidence are studied.Several improved algorithms are presented and applied to multi-sensor multi-target recognition systems.The main research contents are concluded as follows:(1)An improved algorithm of two-two evidence fusion based on D-S combination rule is put forward.In order to fuse the conflict evidence effectively,an improved algorithm for calculating the credibility of focal element is proposed by analyzing invalidation reasons of D-S combination rules at the time of fusing highly conflict pieces of evidence.The improved credibility of focal element is used as the weight of local conflict information redistribution,and a new evidence fusion criterion based on local conflict information redistribution is given.Moreover,the existing local conflict information redistribution algorithm is not satisfied with the associative law which causes the inconstant results of evidence fusion.For overcoming this disadvantage,an improved algorithm based on evidence distance and falsity to measure the evidence conflict is presented,which is used to optimize the fusion sequence of multiple pieces of evidence.And then a two-two evidence fusion algorithm improving the fusion order is proposed.Lastly,the effectiveness of the proposed algorithm is verified by comparing the related algorithms and new methods with numerical examples and target recognition simulations.(2)Direct fusion algorithms of multiple pieces of evidence are presented.Most of the existing evidence fusion algorithms are improved on the basis of two-two evidence fusion.Those improved methods not only lose some useful information because of two-two evidence fusion,but also do not satisfy the associative law,leading to the lack of stability of the algorithm.In order to improve the effect of multiple pieces of evidence fusion in the case of highly conflicting evidence and effectively use the information in the inconsistency factor,an improved D-S combination rule of evidence by directly fusing three pieces of evidence based on local conflict information redistribution is proposed.According to the credibility of each focal element in 3-D conflict factor,conflict information is assigned to focal element involved by weight in the improved target recognition algorithm.On this basis,this algorithm is popularized and applied into the direct fusion of multiple pieces of evidence.At the same time,in order to decrease the computational complexity of the high-dimensional evidence fusion algorithm,and considering that the D-S combination rule has a good focus on non-high-conflict evidence,a selective grouping evidence fusion algorithm is proposed.The new algorithm puts conflict evidence into different groups for fusion,so as to achieve the goal of retaining as much information as possible contained in the multidimensional conflict factor and avoiding destroying the effective use of the D-S combination rules in the inter-group evidence fusion.Theoretical analysis and simulation results show that the new algorithm can improve the probability of target recognition in the case of high conflict evidence in different degrees and has good stability.(3)Two new algorithms of evidence conflict measure are proposed.In the light of the problem of conflict measure of evidence,a local conflict information redistribution algorithm based on evidence ranking fusion is proposed.The new algorithm first measures evidence conflict based on the evidence distance and the conflict coefficient,and then optimizes the sequence of evidence fusion,and improves the conflict measure algorithm of different focal elements in different pieces of evidence.The conflict measure results are low in the case of decentralizing probability assignment in pieces of evidence,so an improved algorithm base on overlapping degree and evidence distance to measure the evidence conflict is developed.This algorithm is not only suitable for the fusion of evidence with single-subset focal element,but also can better solve the fusion between conflicting pieces of evidence containing multi-subsets focal element.