【摘要】 为了解决经典支持向量机方法已发现的执行时间长、执行效率低的相关问题,提出基于增量学习向量的支持向量机学习方法。该算法通过对训练样本集合的相关增量学习向量进行训练学习来得到初始支持向量机分类器。利用该初始化分类器在有关条件下针对初始训练样本集进行缩减得到精简缩小集,再应用精简缩小集针对初始支持向量机的分类器反向加工来得到支持向量机的最终分类器。该算法可大幅度降低大容量数据集上支持向量机的学习时间,并且具有很好的泛化能力。为了验证本学习方法的可应用性,从Berkeley图像分割数据集BSDS500和互联网上选取相关彩色图像进行仿真实验。该文实验结果表明:该方法得到分割结果的过程不仅比传统支持向量机耗时少,且与Berkeley图像分割数据集中人工标注结果比较得到较好分割效果。更多还原

【Abstract】 In order to solve less efficient,longer time-consuming problem of the traditional SVM methods,this paper proposes a support vector machine learning algorithm based on the incremental vector. The algorithm obtains the initial support vector machine classifier by training the sample collection incremental vector learning. This paper streamlines the relevant conditions for initial trainingsample set to be streamlined narrow set by using the initialization classification,applies the thin narrow set of initial support vector machine classifier in reverse processing,and gets the support vector machine classification devices. The algorithm can significantly reduce the training time of support vector machine and large-capacity data set and has good generalization performance. In order to verify the application of the algorithm,this paper selects relative color images from Berkeley image segmentation data set BSDS500 and Internet experiments to do simulation experiment. The experimental results show that this segmentation process has much less time-consuming than the traditional support vector machine and better segmentation than the manually marked results in Berkeley image segmentation dataset.更多还原