基于异态集成学习的飞行目标辅助识别模型  

The Auxiliary Recognition Model for Flying Target Based on Heterogeneous Ensemble Learning

作　　者：刘戎翔 贺筱媛[1] 陶九阳[1,3] LIU Rong-xiang;HE Xiao-yuan;TAO Jiu-yang(Joint Operation College,National Defense University,Beijing 100091,China;Army Chemical Defense Institute of PLA,Beijing 102205,China;Army Engineering University of PLA,Nanjing 210007,China)

机构地区：[1]国防大学联合作战学院,北京100091 [2]陆军防化学院,北京102205 [3]陆军工程大学,南京210007

出　　处：《火力与指挥控制》2020年第4期141-147,共7页Fire Control & Command Control

基　　金：国家自然科学基金面上项目(61403401,61374179,61273189,61174156,61174035);军民共用重大研究计划联合基金资助项目(U1435218)。

摘　　要：对飞行目标类型的准确识别是空中作战意图识别的前提和基础。针对当前各类识别模型在训练样本较少时,较难同时获得模型的稳定性和较好的泛化能力且在线学习能力较差的问题,提出了一种基于异态集成学习的飞行目标辅助识别模型,将k近邻学习模型与BP神经网络模型进行整合,使模型兼具训练稳定性与较好的泛化能力;通过算法设计,模型具有了整体动态更新的能力。基于某作战仿真系统完成飞行目标识别实验,对比了该模型与各类模型的性能表现。实验结果显示所提出的模型识别正确率稳定在90%左右,且在个体学习器的基础上至少提高2%。The accurate identification of the type of Flying target is the prerequisite and basis for the recognition of air combat intention.For the current various types of recognition models,when the training samples are small,it is difficult to obtain the stability and better generalization ability of the model at the same time.Besides,most of the models are lack of online learning ability.This paper proposes anauxiliary flying target recognition model based on heterogeneous ensemble learning,and organic integrate k nearest neighbor learning model and BP neural network model.Make the model has better training stability and generalization ability.Through algorithm design,the model has the ability to update the individual learner and the whole model.Based on that,a flying target recognition experiment is completed on a combat simulation system.The performance of the model is compared with that of single k neighbor learning model and BP neural network model.The experimental results show that the recognition accuracy of flying target recognition model based on heterogeneous ensemble learning is about 90%,and it improves at least 2%based on the accuracy rate of individual learner.

关 键 词：飞行器辅助识别 异态集成学习 k近邻学习模型 BP神经网络模型 模型动态更新 

分 类 号：TJ8[兵器科学与技术—武器系统与运用工程] TP183[自动化与计算机技术—控制理论与控制工程]