摘要
针对雷达目标样本缺乏以及高输入模式维数的分类问题,提出利用一种稀疏概率模型——相关向量机(RVM)对雷达目标的一维距离像进行识别。与支持向量机(SVM)相比,其训练是在贝叶斯框架下进行的,不仅解更稀疏,而且无需调整模型参数。使用RVM与SVM识别同样的雷达目标一维距离像,结果表明:RVM模型更为简单,减少了运算量,但能获得更精确的分类结果。
In view of the problems such as lack of samples and classification of high input dimension, a sparse probabilistic model, termed ＂relevance vector machine＂ （RVM） is proposed for 1-D profile recognition. Compared with support vector machine （SVM）, RVM training based on Bayesian frame, obtain sparser solutions without adjusting any model parameter. The recognition results show that it is much simpler and accurate to recognize the same 1-D range profile with RVM than with SVM.
出处
《探测与控制学报》
CSCD
北大核心
2009年第2期19-21,27,共4页
Journal of Detection & Control
基金
国防基金项目资助(9140A05070107BQ0204)
关键词
目标识别
一维距离像
支持向量机
相关向量机
target recognition
1-Dimage
support vector machine
relevance vector machine
