在所有的训练样本中只有支持向量(SVs)能对支持向量机分界面优化结果产生显著影响.基于是一最近邻规则,提出了一种训练样本的预选取方法.针对一些典型人工数据集、公用基准数据集以及TM遥感数据的实验结果表明,该方法能够有效减少训练样本数日,显著加快学习速度,并保证理想的分类精度.
In support vector machine (SVM) only support vectors (SVs) have the significant influence on the optimization result. An approach for pre-extracting SVs based on k-NN is proposed. The experimental results based on some artificial datasets, some real-world datasets and TM remote sensing dataset show that the approach proposed can effectively reduce the size of training sets and accerlerate the learning speed. At same time, the classification accuracies are ensured.