空中目标作战任务智能判识方法研究  

Research on Intelligent Identification Method of Air Target Combat Task

作　　者：赵文栋 张明智[1] 贺筱媛[1] 郭圣明[1] ZHAO Wen-dong;ZHANG Ming-zhi;HE Xiao-yuan;GUO Sheng-ming(Joint Operations College, National Defence University, Beijing 100091;Unit 61267 of PLA, Beijing 101114, China)

机构地区：[1]国防大学联合作战学院,北京100091 [2]中国人民解放军61267部队,北京101114

出　　处：《指挥控制与仿真》2022年第3期14-19,共6页Command Control & Simulation

基　　金：重大研究计划联合基金项目(U1435218)。

摘　　要：为从动态、时序变化的战场态势中准确迅速地判识出空中目标的作战任务,运用深度学习理论构建作战任务智能判识模型。在门控循环单元网络的基础上,引入双向传播机制构建多层网络,学习作战任务的时序特征;采用残差机制防止出现由于神经网络层数加深导致模型性能退化问题;运用注意力机制,捕捉相互关联的重要态势特征,将先验知识与多层神经网络的输出相融合,确保在获得时序信息较短时对作战任务判识的准确性。实验表明,该方法可以有效地判识空中目标的作战任务。In order to accurately and rapidly identify the combat task from dynamic and time-varying battlefield environment,deep learning theory is used to construct a combat task intelligent identification model.Based on the gated recurrent unit network,the bidirectional propagation mechanism is introduced to build a multi-layer temporal network,and the residual mechanism is adopted to prevent the degradation of model performance caused by the deepening of neural network layers,and then the attention mechanism is applied to capture the interrelated important situation characteristics,and the fusion of prior knowledge and output of the network ensures the accuracy of combat task identification when the temporal information is short.Experimental results show that the proposed method has a good effect.

关 键 词：空中目标 作战任务判识 门控循环单元网络 注意力机制 残差机制 先验知识 

分 类 号：E911[军事] TP391.9[自动化与计算机技术—计算机应用技术]