当支持向量机中存在相互混叠的海量训练样本时,不但支持向量求取困难,且支持向量数目巨大,这两个问题已成为限制其应用的瓶颈问题。该文通过对支持向量几何意义的分析,首先研究了支持向量的分布特性,并提出了基于几何分析的支持向量机快速算法,该算法首先从训练样本中选择出部分近邻向量,然后在进行混叠度分析的基础上,选择真实的边界向量样本子空间用来代替全部训练集,这样既大大减少了训练样本数目,同时去除了混叠严重的奇异样本的影响,并大大减少了支持向量的数目。实验结果表明:该算法在不影响分类性能的条件下,可以加快支持向量机的训练速度和分类速度。

Support vector machine,a research hotspot of the pattern recognition in recent years,performs successfully in solving the nonlinear and high dimensional problems.However,training a support vector machine is equivalent to solving a linearly constrained quadratic programming problem in a number of variables equal to the number of data points.This optimization problem is known to be challenging when existing large number of training data points.Also,it is well known that the number of support vector plays an important role in the classification speed of SVM.So the method of pre-analysis efficient support vectors are used to train classifier becomes a novel task in SVM fields.In this paper,on the basis of a deep investigation into the geometry principle of support vectors and its distribution,we firstly pick out some neighbor vectors by nearest interclass distance analysis,and then select the margin vector by computing its intermixed factor of the neighbor vectors.So this method speeds up the SVM training and classifying synchronously by reducing the number of training samples and trimming the intermixed samples,while the ability of SVM remains unchanged.