边缘效应训练的模糊支持向量机及应用

Edge-effect Training for Fuzzy Support Vector Machine and its application

通常支持向量机算法中每个训练样本的地位是平等的,而实际应用中我们发现边缘训练样本对支持向量机分类性能的贡献大于训练中心区域的样本,为此我们提出一种边缘效应的支持向量机训练算法。在训练样本中增加模糊隶属度属性,从而体现训练样本对分类的不同贡献,突出边缘样本的作用。最后结合卫星图像分割实验,对比证明了新算法的有效性。

In ordinary support vector machine algorithm, all the training samples have equal status, but in practical application we find the edge samples have more important effect on classification results than other samples, so the paper puts forward a kind of edge-effect training algorithm for support vector machine. It gives each training sample a fuzzy membership property, and embodies the different contribution of training samples for classification result and emphasizes the importance of edge samples. At last taking satellite image segmentation as samples, the efficiency of new algorithm is proved.