基于多分类器融合的雷达高分辨距离像目标识别与拒判新方法

A New Method to Improve Radar HRRP Recognition and Outlier Rejection Performances Based on Classifier Combination

由于雷达自动目标识别(Radar automatic target recognition,RATR)中库外目标的存在,评价系统性能时应综合考虑其识别性能和拒判性能.由此本文构造了一种将分类器的输出通过最近邻分类器(Nearest neighbor,NN)进行拒判和识别的"分类器–最近邻"系统,并在拒判和识别两个阶段分别采用多分类器融合技术以提高RATR系统的拒判和识别综合性能.此外,文中定义了一种代价函数以衡量系统综合性能并为系统拒判工作点的选取提供依据.进而,采用局部法和全局法两种算法确定拒判器的工作点.实测数据实验结果验证了本文方法的有效性,两种工作点选取算法均能够显著提高识别系统的综合性能.

Because of the presence of outlier samples, it is important to take both the recognition and rejection perfor- mances into account when evaluating a radar automatic target recognition （RATR） system. In this paper, we propose to utilize the nearest neighbor （NN） classifier with the inputs being the outputs of a classifier, referred to as ＂Classifer-NN＂ system, to identify outliers. In order to improve the performance, several ＂Classifer-NN＂ systems are combined. A cost function is defined to measure the recognition and rejection performance of the RATR system. Two algorithms are devel- oped to select the optimal work point of each ＂Classifer-NN＂ system. Experimental results based on the measured HRRP dataset have validated the effectiveness of the proposed methods.