证据函数的构造方法以及证据推理算法的研究和应用

发布时间：2018-06-04 15:48

本文选题：证据理论 + logistic回归　；参考：《江西师范大学》2017年硕士论文

【摘要】：证据理论的最大优点在于它能在不知道先验概率的前提下表达“不确定性”和“不知道”的问题,为不确定性推理提供了可靠的方法。当前,在人工智能领域已广泛应用。然而,目前还没有从数据源中获取证据函数的完整方法,也就是说,基本信度分配函数的构造和复杂证据网络推理方法仍存在许多问题有待研究。本文的所做工作如下:一、针对证据理论中构造基本信度分配函数(BBA)困难的问题,本文找到了一种加权基本信度指派函数的构造新方法,并应用在多特征图像分类上。该方法以多类logistic回归输出的后验概率与识别正确率构造证据权重系数,进而构造出权重基本信度指派;最后通过加权D-S证据融合最终判别类别。实验结果表明,该方法能够克服单一特征分类精度的不稳定性,提高分类精度。二、本文把团树传播算法应用在证据网络中,解决了复杂的多连通知识网络结构下的信度推理问题。该方法首先把复杂多连通网络构造成一棵团树,并将联合信度作为团节点的参数实现了复杂多连通网络结构上的证据网络信度推理;在进行联合信度函数信息融合过程中,通过引入两种新的交并运算实现了对DSmT组合规则的改进,减少了不确定性。最后,通过一个例子来证明该方法的可行性。
[Abstract]:The greatest advantage of evidence theory lies in its ability to express the problems of "uncertainty" and "not knowing" without knowing the prior probability, which provides a reliable method for uncertain reasoning. At present, it has been widely used in the field of artificial intelligence. However, there is no complete method to obtain the evidence function from the data source, that is, there are still many problems to be studied in the construction of the basic reliability assignment function and the reasoning method of the complex evidence network. The work of this paper is as follows: firstly, in view of the difficulty of constructing the basic reliability assignment function (BBA) in evidence theory, a new method of constructing weighted basic reliability assignment function is found and applied to the classification of multi-feature images. In this method, the weight coefficient of evidence is constructed by using the posterior probability and recognition accuracy of multi-class logistic regression output, and then the basic reliability assignment of weight is constructed. Finally, the final discriminant category is obtained by weighted D-S evidence fusion. The experimental results show that this method can overcome the instability of the classification accuracy of a single feature and improve the classification accuracy. Secondly, the cluster tree propagation algorithm is applied to the evidence network to solve the reliability reasoning problem under the complex multi-connected knowledge network structure. Firstly, the complex multi-connected network is constructed into a cluster tree, and the joint reliability is taken as the parameter of the cluster node to realize the reliability reasoning of the evidential network on the complex multi-connected network structure, and in the process of information fusion of the joint reliability function, Two new intersection and union operations are introduced to improve the DSmT combination rules and reduce the uncertainty. Finally, an example is given to prove the feasibility of the method.
【学位授予单位】：江西师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】