摘要
本文针对支持向量机难以快速有效地进行增量式学习的问题,提出了一种基于内壳向量的支持向量机增量式学习算法。算法通过线性规划运算求得最可能包含支持向量的壳向量和内壳向量集合,在保证分类精度的前提下最大程度地缩小训练集规模,进而在新的训练集中快速训练支持向量机。将该算法应用于公开数据及低空飞行声目标分类识别,结果表明,新算法缩短了训练时间,且比现有其他算法具备更高的分类精度和稳定性。
We present an algorithm based on inner hull vectors for SVM incremental learning in this paper. In our algorithm, a set of hull vectors and inner hull vectors which most likely to become the support vectors are extracted from the training samples by using the linear programming, the obtained hull vectors and inner hull vectors are conjoined as a part of updated training samples which is smaller than the original training samples, then using the updated training samples to reconstruct the SVM. The proposed algorithm is tested on public databases and low altitude tlying acoustic targets data. Experiment results show that the proposed method is more precise and stable than the other methods and also expedite the SVM training.
出处
《电路与系统学报》
CSCD
北大核心
2011年第6期109-113,共5页
Journal of Circuits and Systems
基金
国家自然科学基金(60872113)
关键词
支持向量机
增量学习
壳向量
内壳向量
support vector machine
incremental learning
hull vector
inner hull vecltors
