加权KNN分类器在HRRP库外目标拒判中的应用

Application of a weighted KNN classifier for HRRP out-of-database target rejection

针对雷达自动目标识别中的库外目标拒判问题,提出了一种人工生成库外样本的方法和一种加权k最邻近(knearest neighbors,KNN)分类器。通过人工生成库外高分辨距离像样本,解决了在训练阶段无法获取库外样本的难题。加权KNN分类器同时满足了基于问题和基于数据两大设计要求,能够很好地处理拒判问题。通过基于接收机工作特性(receiver operating characteristic,ROC)准则和基于损失函数准则的仿真实验,证明了加权KNN分类器具备优良的拒判性能。

To cope with the out-of-database target rejection problem in radar automatic target recognition（ATR）,a method that artificially generates out-of-database examples and a weighted k nearest neighbors（KNN） classifier are proposed.By artificially generating out-of-database high-resolution range profiles（HRRPs）,the problem of acquiring out-of-database examples during the training step is solved.The weighted KNN classifier is both problem-dependent and data-dependent,and thus it is well suited for the associated rejection problem.Experiments conducted under the receiver operating characteristic（ROC） rule and the loss function rule respectively show that the weighted KNN classifier obtains satisfactory rejection ability.