基于子类划分的SVM及水声目标识别应用

SVM Based on Subclass and Application of Underwater Acoustic Target Recognition

针对目标类空间的可分离性特点,研究了动态聚类与支持向量机相结合的基于子类划分的支持向量机。提出了以子类中心为基点度量训练样本惩罚度的方法。在采用动态聚类将目标类划分为子类的基础上,综合考虑训练样本与所属子类的距离、子类对所属目标类的隶属度及目标类间的关系,以度量不同训练样本的惩罚度。并应用于水声目标识别中,对两类舰船目标的识别情况进行了比较,实验表明效果好于经典支持向量机。

In view of separability trait of objective class space to highly nonlinear problem,the support vector machine based on subclass integrating the dynamic clustering with the support vector machine （SVM） was researched.A method was put forward that measured the penalization-degree using the subclass center as base point.On the basis of objective class decomposed as subclasses by using dynamic clustering,the penalization degree of each sample was measured by synthetically considering the relation between each sample and its subclass,the relation between the subclass and its objective class and the relation in different objective classes.At last,the SVM based on subclass was applied in the underwater acoustic target recognition.Experiment results show that the SVM based on subclass is more robust than the traditional SVM.