%0 Journal Article
%T 最短特征线段多分类器系统设计<br>Design of Multiple Classifier Systems Based on Shortest Feature Line Segment
%A 丁建坤
%A 韩德强
%A 杨艺
%J 西安交通大学学报
%D 2015
%R 10.7652/xjtuxb201509014
%X 为提高多分类器系统分类的性能,设计了一种使用最短特征线段分类器的多分类器系统。依据最短特征线段分类算法工作机理,利用特征线段长度表征样本隶属于各个类别的可能性,即模糊隶属度,对成员分类器输出形式完成由摘要级至度量级的重新建模,更多地保留输出细节以减少信息损失,进而利用基于模糊的证据融合规则实现成员分类器的度量级融合,通过隶属度到mass函数的转换,利用模糊？仓ぞ萑诤瞎嬖蚴迪侄喾掷嗥飨低车墓乖？,进一步提高了多分类器系统分类性能。采用人工数据集和UCI数据集设计了对比实验,实验表明,与其他邻域型分类器构造的多分类器系统相比,新多分类器系统能有效提升分类正确率。<br>To improve the classification performance of multiple classifier systems, a novel multiple classifier system using shortest feature line segment (SFLS) as member classifiers is proposed. According to the SFLS’s algorithmic principle on classification, the length of the shortest feature line segment is used to represent the probability, i.e., the fuzzy membership of the query sample belonging to the corresponding class. Thus, the member classifier’s output is transformed from the abstract level to the measurement level. Furthermore, by transforming the fuzzy membership function into the mass function and using fuzzy？？based evidential fusion rules, the classification fusion is implemented. Compared with traditional multiple classifier systems, the proposed approach can use more detailed information for implementing more effective decision？？level fusion. Experimental results show that the proposed multiple classifier systems can effectively improve classification accuracy
%K 特征线段
%K 隶属度
%K 多分类器系统
%K 证据理论&lt
%K br&gt
%K feature line segment
%K membership function
%K multiple classifier systems
%K evidence theory
%U http://zkxb.xjtu.edu.cn/oa/DArticle.aspx?type=view&id=201509014