摘要
针对当前SVDD算法由于过大的优化规模导致检测计算时间过长的问题,提出了动态SVDD算法。通过分析在进行检测工作时新加入检测对象对正域边界的影响,提出:采用核方法形成的边界可近似替代折线所形成的边界。这样,加入新检测对象后,新的边界就只与新的样本点和之前的边界有关,从而可以大大减小优化规模,提高检测的效率。
In order to solve the problem of long computation time in detecting caused by over-large optimization scale in SVDD,a dynamic support vector data description was proposed. After analyzing a new object' s influence on positive border,it was suggested that the boundary formed by kernel methods could be approximately replaced by boundary formed by polygonal lines. Thus, after adding new objects, the corresponding new boundary was only related with new objects and previous boundary, which means the optimization scale was largely decreased and the efficiency of detecting was promoted.
出处
《计算机科学》
CSCD
北大核心
2009年第3期156-157,183,共3页
Computer Science
基金
中国博士后科学基金资助项目(2005038042)
广东省科技计划项目(2006B12701002)资助
关键词
SVDD
边界
支持向量
核方法
优化规模
SVDD,Boundary,Support vector,Kernel methods,Optimization scale