一种基于Logistic回归的基本信度分配函数的构造新方法

The New Method Based on Logistic Regression of Constructing Basic Belief Assignment Function

结合Logistic回归分类,该文提出一种新的构造证据理论基本信度分配函数的方法,并将其应用于多特征图像分类.该方法首先以多类Logistic回归分类法输出的后验概率与样本分类正确率建立证据权重系数,然后构造出加权的基本信度分配函数,最后利用加权D-S证据融合判别所属类别.实验结果显示:该方法既能提高图像分类的正确率,又能改正使用单特征分类导致的分类正确率的不稳定的缺点.

Combined with Logistic regression classification,the new method of constructing the basic belief assignment function of evidence theory is presented,and it is applied in the multi-feature image classification.Firstly,the weight coefficient of the evidence is established based on the posterior probability output by multi-class Logistic regression classification method and the samples′classification accuracy.Secondly,the weighted basic belief assignment function is constructed.Finally,the weighted D-S evidence fusion is used to distinguish the category.Experimental results show that the new method can not only improve the accuracy of image classification,but also overcome the instability of classification accuracy caused by single feature classification.