基于动态自适应近邻算法的天波雷达RD图分类器设计  

RD Image Classifier Design Based on Dynamic Adaptive Nearest Neighbor Algorithm for Sky Wave Radar

作　　者：罗忠涛 唐洪涛 高天翱 曹健 LUO Zhongtao;TANG Hongtao;GAO Tianao;CAO Jian(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Nanjing Research Institute of Electronics Technology,Nanjing 210013,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]南京电子技术研究所,南京210013

出　　处：《电讯技术》2024年第8期1315-1321,共7页Telecommunication Engineering

基　　金：国家自然科学基金资助项目(61701067,61702065)。

摘　　要：天波雷达的干扰检测问题可转化为距离-多普勒(Range-Doppler, RD)图像分类。在RD图分类器设计中,使用K近邻(K-Nearest Neighbor, KNN)算法时,k值的选取直接影响到干扰检测准确率。根据过往经验预设k值时,无法确定所设k值下的干扰检测准确率。为此,将互近邻条件引申为k值自动赋值方法,以局部均值为距离计算依据,设计动态自适应近邻(Dynamic Adaptive Nearest Neighbor, DANN)新算法。分别在多个UCI(University of California Irvine)数据集与现有RD图库上测试,与6个常数k值下K近邻算法进行对比分析。多个UCI数据集实验表明,DANN的平均准确率比不同k值下KNN的均值高6.21%,且比最优k值高3.7%;实测RD图库实验表明,DANN的平均准确率比不同k值下KNN的均值高2.9%,且比最优k值高0.56%。因此,该算法可以在干扰检测中减少人工参与,且能够获得较高的检测准确率。The interference detection problem of sky wave radar can be converted into range-Doppler(RD)image classification.When K-nearest neighbor(KNN)algorithm is used in the design of RD image classifier,the selection of k value directly affects the accuracy of interference detection.The k values is required to be preset according to past experience,but the accuracy of interference detection cannot be anticipated under the set k values.Therefore,the authors extend the mutual nearest neighbor condition to the k value automatic assignment method,and by using the local mean as the distance calculation basis,designs a new dynamic adaptive nearest neighbor(DANN)algorithm.The proposed algorithm is testel on multiple University of Calitornia Irvine(UCI)datasets and existing RD image dataset,and compared with the KNN algorithm with 6 constant k values.Multiple UCI dataset experiments show that the average accuracy of DANN is 6.21%higher than the mean of KNN under different k values,and 3.7%higher than the optimal k value.The real RD image dataset experiment shows that the average accuracy of DANN is 2.9%higher than the mean of KNN under different k values,and 0.56%higher than the optimal k value.The proposed algorithm can reduce manual involvement in interference detection and obtain a better detection accuracy.

关 键 词：天波雷达 干扰检测 RD图像分类 自适应近邻 

分 类 号：TN958.93[电子电信—信号与信息处理]