摘要
针对遥感全脉冲数据“大数据、小样本”,传统基于模板序列匹配的识别方法时间复杂度高,深度学习算法训练样本不足导致泛化能力差的问题,提出一种基于AdaBoost.M2-DT算法的识别方法。首先采用决策树实现基分类器,然后使用Adaboost.M2的集成学习方法构建识别模型。实验采用9型雷达辐射源的外场数据,分别使用序列匹配方法、SVM(支持向量机)、CNN(卷积神经网络)和AdaBoost.M2-DT算法进行训练和识别并对比实验结果,表明Ada-Boost.M2-DT算法对小样本的遥感全脉冲数据具有较高的识别正确率和较小的时间复杂度。
Because of remote sensing full pulse which has a large amount of data,and few highquality samples,the traditional recognition method based on template sequence matching has high time complexity,the deep learning algorithm has poor generalization ability.To solve this problem,AdaBoost.M2-DT algorithm is proposed.In the first step,the decision tree is used to implement the classifier.The second step is to build the recognition model with AdaBoost.M2 algorithm.In the experiment,the outfield data of 9 kinds of radar emitter are used,the sequence matching method,SVM,CNN and AdaBoost.M2-DT algorithm are used to train,identify,and compare the experiment results.It shows that AdaBoost.M2-DT algorithm has higher recogni-tion accuracy and less time complexity for small samples of remote sensing full pulse data.
作者
雷涛
王丽军
徐晶
毕晓伟
LEI Tao;WANG Lijun;XU Jing;BI Xiaowei(Southwest China Research Institute of Electronic Equipment,Chengdu 610036,China)
出处
《电子信息对抗技术》
2020年第5期6-10,41,共6页
Electronic Information Warfare Technology
关键词
辐射源型号识别
遥感
全脉冲
小样本
radar emitter type identification
remote sensing
full-pulse data
small sample
