一种基于集成学习的战术目标类别识别方法  

A Tactical Target Category Recognition Method Based on Integrated Learning

作　　者：戎纪光[1] 刘伟刚 田聚波[1] 赵鹏[1] RONG Jiguang;LIU Weigang;TIAN Jubo;ZHAO Peng(The 54th Research Institute of CETC,Shijiazhuang 050081,China)

机构地区：[1]中国电子科技集团公司第五十四研究所,河北石家庄050081

出　　处：《无线电工程》2023年第7期1700-1705,共6页Radio Engineering

基　　金：国家自然科学基金青年基金项目(62101517)。

摘　　要：针对实际运用中的战术目标类别识别精度不高的问题,设计了一种战术目标类别识别框架,提出了一种基于Borderline-SMOTE的改进小样本战术目标数据采样方法,利用战术目标特征个体在整体数据集中的互信息的sigmoid函数作为遗传算法的适应度函数,并运用到选择、交叉和终止计算中,从而加速最优特征子集的选择,使用LightGBM等算法搭建一个战术目标类别的识别的Stacking集成算法模型,通过历史数据验证了该方法的有效性。In view of the low accuracy of tactical target category recognition in practical applications,a framework for tactical target category recognition is designed,and an improved small-sample tactical target data sampling method based on Borderline-SMOTE is proposed.The sigmod function of the mutual information of tactical target feature individuals in the overall data set is used as the fitness function of genetic algorithm,and is applied to selection,crossover and termination calculations,so as to accelerate the selection of the optimal feature subset.LightGBM and other algorithms are used to build a stacking integrated algorithm model for the recognition of tactical target categories,and finally the effectiveness of this method is verified through historical data.

关 键 词：战术目标类别识别 小样本采样 特征选择 集成学习 基分类器 元分类器 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]