中文摘要：

针对雷达高分辨距离像的姿态敏感性，利用各目标在各种姿态下的雷达高分辨距离像样本和其局部聚散程度信息，调节各样本的局部有效作用范围，使得其统计置信水平达到可信的范围内，从而对二分类的K近邻测度距离和判决准则进行了优化，最后通过一对一法将其推广到解决多类目标的识别问题．实验证明该方法相对于传统的K近邻方法可有效提高识别率，尤其当类别增多时，性能改善显著．

英文摘要：

Due to the target-aspect sensitivity of high resolution rang profiles （HRRPs）, we first use a modified k-nearest neighbor （KNN） rule for binary-class classification problem, and then extend it to multi-class classification problems using the one against one （OAO） method. This method adjusts the effective influence size of each training sample in order to make sure the statistical confidence level in a range that can be trusted. Experimental results show our method＇s good performance for multi-class classification problems and its effectiveness to improve the KNN rule.