论文部分内容阅读
为进一步提高多分类器系统的分类性能,提出了一种基于知识发现的特征决策层多分类器融合新方法.各分类器工作于具有互补分类信息的不同特征空间且其类型由不同的类间可分性度量决定.各分类器输出的不确定性度量从建立的多个决策表中导出,并具有条件mass函数的形式.进而基于广义粗集模型和Dempster-Shafer理论(DST)构造了一种新颖的特征决策层融合框架.高光谱遥感图像的分类实验表明,与多数表决融合(PV)相比,所提出的方法可有效提高多分类器系统的分类性能.
In order to further improve the classification performance of multi-classifier systems, a new fusion method based on knowledge discovery for multi-classifier of feature decision-making layers is proposed.Each classifier works in different feature spaces with complementary classification information and its type consists of different classes Separability metric.Uncertainty metrics output from each classifier are derived from the established multiple decision tables and have the form of conditional mass functions.Then a generalized rough set model and Dempster-Shafer theory (DST) A novel feature-based decision-making fusion framework was proposed.The classification experiment of hyperspectral remote sensing images showed that the proposed method can effectively improve the classification performance of multi-classifier systems compared with most voting fusion (PV).