基于LightGBM与特征工程的雷达辐射源识别方法  

Radar Emitter Identification Based on LightGBM and Feature Engineering

作　　者：陈歆普 欧旺军 张智 CHEN Xinpu;OU Wangjun;ZHANG Zhi(Science and Technology on Electronic Information Control Laboratory,Chengdu 610036,China)

机构地区：[1]电子信息控制重点实验室,成都610036

出　　处：《电子信息对抗技术》2021年第5期54-58,共5页Electronic Information Warfare Technology

摘　　要：针对雷达辐射源型号识别,提出一种基于LightGBM集成树学习与特征工程的识别方法。识别技术采用模式识别方法,由特征提取和分类器组成。信号参数级的分选结果是一种表格式的结构化数据。采用基于LightGBM集成树的分类算法,对于结构化数据,该算法展现出优异的性能。利用特征工程对分选结果中的数值属性和标称属性进行转换,作为LightGBM学习算法的输入特征。测试结果表明LightGBM算法在仿真数据集上能够有效识别雷达辐射源。基于LightGBM集成树模型,数值特征和标称特征的融合能够近一步提升识别效果。For radar emitter identification,a recognition approach based on LightGBM ensemble tree learning and feature engineering is proposed.The identification technique is based on pat-tern recognition,consisted of feature extraction and classifier algorithms.The sorting result on signal-parameter level is a type of tabular structured data.The LightGBM ensemble tree learn-ing,which has achieved state-of-art performances on the structured data,is chosen as a classifi-er algorithm.Feature engineering approaches are applied to converting sorting results into fea-tures for the LightGBM input,with numeric and nominal attributes.Testing results show that the LightGBM algorithm is able to recognize radar emitters on the simulation data set.Based on the LightGBM ensemble tree model,the recognition accuracy could be improved further with the combination of numeric and nominal features.

关 键 词：雷达辐射源型号识别 模式识别 LightGBM集成树学习 特征工程 

分 类 号：TN971.1[电子电信—信号与信息处理]