基于高斯混合模型的辐射源模式识别算法  

Gaussian Mixture Model Based Algorithm for Radiator Pattern Recognition

作　　者：栗大鹏 梁伟 LI Dapeng;LIANG Wei(School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China;Beijing Institute of Remote Sensing & Equipment, Beijing 100854, China)

机构地区：[1]北京理工大学机电学院,北京100081 [2]北京遥感设备研究所,北京100854

出　　处：《探测与控制学报》2017年第6期40-45,共6页Journal of Detection & Control

基　　金：国防973计划项目资助(616196)

摘　　要：针对现有算法对战场新出现辐射源学习与分类能力较差的问题,提出了基于高斯混合模型(Gaussian Mixture Model,GMM)的辐射源信号模式识别算法,该算法对信号在特征空间中的分布模式进行在线学习,形成基于概率统计的分类模型,在分类中给出样本归属的似然概率。为了进一步提高算法的实时性,提出基于空间网格划分的快速EM(Expectation Maximization)方法,从而使GMM拟合的计算复杂度取决于网格划分的密度而不是样本数量,极大提高了算法效率。对电子侦察典型场景的仿真分析表明,算法能够对各类辐射源进行在线学习与分类,适应性较强,且计算效率较传统EM算法有较大提高。To overcome the difficulties of learning and classifying newly emerged signal patterns,a Gaussian mixture model(GMM)based algorithm for radiator pattern recognition was proposed.The algorithm could perform online learning on the signal's distribution in the feature space,which formed a probabilitybased classification model which classified samples by likelihood.Besides,to match the realtime requirement of the system,a gridbased fast expectation maximization(EM)method was proposed,which made the caculation complexity of GMM fitting be propotional to the grid amount but not to the sample amount.Computer simulations showed that the proposed algorithm was capable of learning and classifying new signal patterns and the computational efficency had greatly improved compared with traditional EM method.

关 键 词：高斯混合模型 模式学习 模式分类 EM算法 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]