类空间规整度的计算几何组合分类器权重分配

在计算几何组合分类器中,子分类器的权重分配一直未能充分利用空间视觉信息,使得分类器的可视化特性无法完全得到发挥.本文从类空间类别分布特性出发,提出基于类空间规整度的权重分配方法.该方法首先将子分类器由空间的类别表示转变为类别的空间表示,进而利用共生原则分析不同类别在空间中的分布规整度.由于分布规整度为类别分布信息的整体体现,可以用于刻画类空间中不同类别样本的离散程度,因此可以利用当前类空间的规整度信息作为该子分类器的权重.实验表明,利用规整度信息进行加权后的分类器不但与可视化特性更好的吻合,增强了分类过程的可理解性,而且在分类精度上得到了进一步的提升,扩展了应用领域.

Weight Calculation for Computational Geometry Combining Classifier Using Regularity of Class Space

In all the tissues about computational geometry combining classifier,the weight calculation for sub classifiers has not taken the advantage of visual information in the spaces,which retains the visual performance about classifier.According to the category distribution in class space,a weight calculation method based on space regulation is proposed.In this method,the space is turned from category information in space to space information in category.And the space regularity is obtained from the later based on co occur rules.As the regularity reflects the distribution of categories and describes the separation of the samples,which makes it as the weight for the sub classifier.The experiments show that the classifier weighted by the regularity not only enhance the visual performance,but also the classify performance of the classifier.It means that the comprehensibility of the classifier is enhanced and the application of the classifier is extended.