摘要
模糊支持向量机是在不可分样本集情况下进行模式分类的有效工具,为了进一步提高该算法的推广能力,对其进行了两方面的改进。一是在高维特征空间中引入不等距分类超平面,以期提高该算法的学习精度;二是在高维特征空间中,利用本文所提出的算法,筛选出有效的训练样本集,以期缩短该算法学习所耗时间。对模糊支持向量机的改进进行了理论推导,并且给出了有效训练样本集的筛选算法。把上述改进方案应用到两种飞机的雷达一维距离像识别中,实验结果表明其取得了很好的识别效果,并且缩短了算法学习时间。
Fuzzy support vector machines （F-SVM） algorithm is effective for pattern classification on unclassifiable sample sets condition. For the sake of enhancing the algorithm＇s applicability, it was im- proved in this paper from two aspects. One hand, non-equidistant margin hyperplane （NM） in high di- mension feature space is introduced to improve on study precision; On the other hand, effectual training sample sets in high dimension feature space are filtrated, via algorithm introduced by this paper, to re- duce study time. This paper gave the theoretical derivation of improved F-SVM and the filter algorithm of effectual training sample sets. The improved methods were applied to Radar Range Profile＇s Recogni- tion of two planes. Experimental results show that these methods can obtain very excellent recognition effect and reduce the algorithm study time.
出处
《计算机工程与科学》
CSCD
北大核心
2013年第6期82-87,共6页
Computer Engineering & Science
基金
人工智能四川省重点实验室项目(2011RYJ06
2011RYY05)
关键词
模糊支持向量机
不等距分类超平面
特征空间
雷达一维距离像
fuzzy support vector machines
non-equidistant margin hyperplane
feature space
radarrange profile
