【摘要】 针对单一整体核函数相对于高维原始空间特征局部分布缺乏自适应性,导致特征利用效率低、分类性能下降明显的问题,提出分解子空间自适应核函数综合支持向量机解决思路.首先将原始空间的高维特征分成一系列有意义的低维子空间;然后针对不同子空间选择不同的核函数和核参数,根据子空间SVM分类性能进行置信度加权;最后,综合所有有效子空间形成的综合核矩阵构造综合支持向量机.由于不同输入子空间构造的核函数具有更大灵活性,综合核矩阵较好利用了高效子空间特征的分类能力.实验结果表明,该方法能够较好地改进高维模式识别的性能,是一种值得实际应用推广的方法.更多还原

【Abstract】 Unitary kernel function is deficient in compatible with the feature distribution especially when the input space is high dimension,and it results in the performance of classification becoming worse.In order to solve this problem,an enhanced classical support vector machine(SVM) based on integrated local subspace adaptive kernel function is proposed.Firstly the input space is divided into a series of significant low dimensional subspace;then the different kernel parameters are obtained by constructing a series of subspace SVM classifier;finally the integrated kernel matrix and enhanced SVM is setup by using all the information of subspace kernel function.Because integrated kernel matrix uses all the classification ability provided by the significant subspace,it performs better than the general unitary kernel SVM.Compared experimental results show that this algorithm performs better in high dimensional pattern recognition application,and may be applicable in many other fields.更多还原