一种基于Dirichelt过程隐变量支撑向量机模型的目标识别方法

发布时间：2018-03-29 07:34

本文选题：目标识别　切入点：混合专家系统　出处：《电子与信息学报》2015年01期

【摘要】：在目标识别中,对于样本数较多且分布复杂的数据,若将所有训练样本用来训练一个单一的分类器,会增加分类器的训练复杂度,且容易忽视样本的内在结构,不利于分类。因此人们提出了混合专家系统(ME),即将训练样本集划分为多个训练样本子集,并在每个子集上单独训练分类器。但是传统ME系统需要人为确定专家个数,并且每个子集的学习独立于后端的任务,如分类。该文提出一种基于Dirichlet过程(DP)混合隐变量(LV)支持向量机(SVM)模型(DPLVSVM)的目标识别算法,采用DP混合模型自动确定样本聚类个数,同时每个聚类中使用线性隐变量SVM(LVSVM)进行分类。不同于以往算法,DPLVSVM将聚类过程和分类器的训练过程联合优化,保证了各个子集中样本的分布上的一致性和可分性,而且可以利用Gibbs采样技术对模型参数进行简便有效的估计。基于人工数据集、公共数据集以及雷达实测数据的实验验证了该文方法的有效性。
[Abstract]:In target recognition, if all training samples are used to train a single classifier, the training complexity of the classifier is increased, and the intrinsic structure of the sample is easily ignored. Therefore, a hybrid expert system is proposed, that is, the training sample set is divided into several subsets of training samples, and the classifier is trained separately on each subset, but the traditional ME system needs to determine the number of experts artificially. Moreover, the learning of each subset is independent of back-end tasks, such as classification. In this paper, a target recognition algorithm based on the Dirichlet process and the mixture of hidden variables (Dirichlet) and support vector machine (SVM) model is proposed. The DPLVSVM) algorithm is used to automatically determine the number of samples by using the DP hybrid model. At the same time, each cluster is classified by using the linear hidden variable SVM- LVSVM. Different from the previous algorithm, DPLV SVM optimizes the clustering process and the training process of the classifier, which ensures the consistency and separability of the distribution of samples in each subset. Moreover, the model parameters can be easily and effectively estimated by using Gibbs sampling technology, and the effectiveness of this method is verified by experiments based on artificial data sets, common data sets and radar measured data.
【作者单位】：西安电子科技大学雷达信号处理国家重点实验室;
【基金】：国家自然科学基金(61372132,61271024,61322103) 新世纪优秀人才支持计划(NCET-13-0945) 全国优秀博士学位论文作者专项资金(FANEDD-201156) 中央高校基本科研业务费专项资金资助课题
【分类号】：TN957.52;TP181

【相似文献】