DSmT与DST融合门限改进方法

来源:网络(转载) 作者:刘永阔 凌霜寒 发表于:2012-04-16 12:11  点击:
【关健词】冲突距离函数;DSmT;DST;门限;信息融合
Dezert.Smarandache理论(DSmT)是一种能够高效实现多源信息融合,成功处理强冲突证据源的数据融合方法,而Dempster.Shafer理论(DST)在证据源冲突低时的融合效果好,运算代价低。将两种技术结合,在冲突距离函数变化率较低时采取DST证据理论,反之采用DSmT融合算法是

 Threshold improvement method combining DSmT and DST
  
  
  LIU Yong.kuo1,2,LING Shuang.han1*
  1. College of Nuclear Science and Technology, Harbin Engineering University, Harbin Helongjiang 150001, China;
  2. National Key Laboratory of Reactor System Design Technology, Nuclear Power Institute of China,Chengdu Sichuan 610041, China
  
  Abstract:
  DSmT is a data fusion method, in which high conflicting evidence sources could be successfully handled,to efficiently realize multi-source information fusion. Meanwhile, Dempster-Shafer Theory (DST) can bring a better result with less computational cost on condition that conflicts are low. Therefore,the integration of the two methods,which is,the DST evidence theory will be adopted when the conflicts are lower, otherwise the DSmT fusion algorithms will be used, is a feasible way to raise the efficiency of the information fusion. The method of single-value switching thresholds for DSmT and DST has been proposed. According to the deficiency of the method,this article proposes that the conflict distance function can be regarded as the judgement basis. Thus the single-value thresholds and the multi-spot value thresholds are distinguishable according to different evidence combinations.
  
  Dezert.Smarandache Theory (DSmT) is a data fusion method, in which high conflicting evidence sources could be successfully handled, to efficiently realize multi.source information fusion. Meanwhile, Dempster.Shafer Theory (DST) can bring a better result with less computational cost on condition that conflicts are low. Therefore, integrating the two methods, the DST evidence theory will be adopted when the conflicts are lower, otherwise the Dezert.Smarandache Theory (DSmT) fusion algorithms will be used, which is a feasible way to raise the efficiency of the information fusion. The method of single.value switching thresholds for DSmT and DST has been proposed. According to the deficiency of the method,this article proposed that the conflict distance function can be regarded as the judgment basis. Thus, the single.value thresholds and the multi.spot value thresholds are distinguishable according to different evidence combinations.
  
  Key words:
  conflict and distance function; Dezert.Smarandache Theory (DSmT); Dempster.Shafer Theory (DST);threshold; information fusion
  
  
  
  0 引言
  D.S理论(Dempster.Shafer Theory,DST),在信息融合,特别是在决策级融合中已经得到了广泛应用。但在证据源存在高冲突的情况下,D.S证据理论会得出许多悖论,许多学者就这类问题进行了研究[1~3]。DSmT(Dezert.Smarandache Theory)作为证据理论的一种发展和延伸,能够在高冲突的情况下得到比D.S证据理论合理的融合结果,但是,DSmT和DST相比,计算量和存储量大,在低冲突的情况下融合效果不及DST[4]。因此,希望能够在实际运用时将两者结合使用,使两种理论实现优势互补——即能在信息源冲突低的情况下以牺牲较小的计算代价获得好的融合效果,又能高效地完成对强冲突信息的融合。在文献[4]的基础上本文就这两种方法的综合使用,提出了新的转换门限确定方法,为信息的融合提供了新的思路。 1 DST与DSmT组合规则
  DST能够处理不确定的信息。作为一种不确定推理方法,其特点是:可满足比贝叶斯概率论更弱的条件;具有表达“不知道”和“不确定”的能力。DST是建立在一个非空集合Θ上的理论[5],Shafer称其为识别框架。设m1(A)和m2(B)是同一个识别框架Θ上基于两个独立证据的基本概率赋值函数,焦元分别是A1,A2,…,Aj和B1,B2,…,Bj。经典DST组合准则[6]:
  
  m(C) = ∑Ai ∩Bj = C m1 (Ai)m2 (Bj )1-k,CΘ,C≠0,C= (1)
  其中:
  k = ∑Ai ∩Bi = m1 (Ai )m2 (Bj ) <1(2)
  DSmT是由法国学者Dezert在2002年提出来的,是经典证据理论的延伸。DSmT能够融合任何类型的独立的信息或证据,但是主要集中在融合不确定、高冲突、不精确的信息源。可以跳出DST框架的局限解决复杂的静态或动态融合问题[7~8]。
  
  
  假设同一识别框架Θ={θ1,θ2,…,θi}下的多条独立的、不确定的和荒谬的(即高冲突的)证据源,和定义在DΘ(或DΘ的任何子集)上的多个广义基本概率赋值函数m1(·),m2(·),…,mk(·),混合式DSmT组合准则为:
  
  摘要:Dezert.Smarandache理论(DSmT)是一种能够高效实现多源信息融合,成功处理强冲突证据源的数据融合方法,而Dempster.Shafer理论(DST)在证据源冲突低时的融合效果好,运算代价低。将两种技术结合,在冲突距离函数变化率较低时采取DST证据理论,反之采用DSmT融合算法是一种提高信息融合效率的可行方式。研究人员对DSmT和DST二者的单点值转换门限方法已做了探讨,针对单点值门限方法的不足,提出了将冲突距离函数作为判别依据来确定转换门限的方法。该方法有很强的适应性,根据不同的证据组合,能划分是单点值门限还是多点值门限。
  
  关键词:冲突距离函数;DSmT;DST;门限;信息融合
  
  中图分类号: TP274.2 文献标志码:A
  
  Threshold improvement method combining DSmT and DST (责任编辑:南粤论文中心)转贴于南粤论文中心: http://www.nylw.net(南粤论文中心__代写代发论文_毕业论文带写_广州职称论文代发_广州论文网)

顶一下
(0)
0%
踩一下
(0)
0%


版权声明:因本文均来自于网络,如果有版权方面侵犯,请及时联系本站删除.