一种基于半监督学习的弹道目标识别方法  

A Ballistic Target Recognition Method Based on Semi-Supervised Learning

作　　者：田西兰[1,2] 郭法滨 赵洪立[1,2] 

机构地区：[1]中国电子科技集团公司第三十八研究所,安徽合肥230088 [2]孔径阵列与空间探测安徽省重点实验室,安徽合肥230088

出　　处：《雷达科学与技术》2017年第6期651-655,共5页Radar Science and Technology

摘　　要：常规反导目标识别系统多依赖于先验知识与一定规模的样本,然而,由于弹道导弹发射事件具有偶发性强、非合作性强等特点,弹道目标样本积累困难,已积累的观测数据也难以进行准确的类别标定。因此,弹道目标识别系统通常研发周期较长、开发代价巨大。针对该问题,将半监督学习算法引入弹道目标识别分类器设计,以降低常规分类识别方法对样本规模的要求。进一步地,针对弹道目标特征分布未知的情况,设计一种TSVM-MKL分类器实现对半监督学习中假设模型的自适应调整。数据验证结果表明,该算法能在"极小"标定样本识别情景下仍能取得良好的分类识别效果,具有良好的工程应用价值。Classical ballistic target recognition system is based on prior knowledge and a large scale sample set.However,it is difficult to set up such data set because the ballistic missile launching is rare and noncooperative.Also,it is hard to make classification based on the measured data already obtained.The abovementioned issues result in a long and high-cost development cycle.Different from the classical classification algorithm which requires a mass of samples,the semi-supervised learning is introduced into the design of classifier for ballistic target recognition to reduce the sample scale.Further,for the case of unknown character distribution of ballistic target,a TSVM-MKL classifier is designed to perform adaptive adjustment of the unknown model assumption in semi-supervised learning.Data analysis results demonstrate that the proposed method works well under rare labeled sample data and is suitable for practical use.

关 键 词：半监督学习 弹道目标识别 多核学习 多核直推式支持向量机(TSVM-MKL) 

分 类 号：TN958[电子电信—信号与信息处理] TJ765[电子电信—信息与通信工程]