摘要
分析了弹道目标识别的特点,指出弹道真假目标识别在本质上是一个风险最小的寻优过程,等同于Neyman-Pearson决策。在此基础上,提出采用ROC曲线下的面积(AUC)对识别方法进行评估,介绍了AUC的两种计算方法。进一步,提出采用AUC对识别器进行比较及优化,介绍了ROC凸包的基本概念并将其用于识别器优化。进行了相应的仿真实验,结果表明了所提出方法的可行性。
After the characteristic of ballistic target recognition was analyzed, it was pointed that alleging decoys was a Neyman-Pearson decision in essence, which sought the least cost of recognition. The area under receiver operating characteristic （AUC） was proposed to use for assessing recognition approach and two methods of calculating AUC were introduced. Furthermore, the AUC were put forward for comparing and optimizing classifier. The ROC convex hull was introduced and used for optimizing classifier. The relevant simulation results verify the effectiveness of the proposed method.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2007年第17期4028-4032,4054,共6页
Journal of System Simulation
基金
国家自然科学基金资助项目(60672033)
新世纪优秀人才支持计划项目(NCET-04-0997)
关键词
弹道目标
识别
评估
接收机工作曲线
ballistic target
recognition
assessment
receiver operating characteristic （ROC）
