A hierarchical aggregation model for combat intention recognition

一种识别作战意图的层次聚合模型

¹ School of Computer Science, Northwestern Polytechnical University, Xi'an 710072, China
² School of Software, Northwestern Polytechnical University, Xi'an 710072, China

Received: 10 June 2022

Abstract

Combat intent recognition refers to analyzing the enemy target's state information to interpret and judge the purpose of the enemy. With the increased knowledge of combat platforms, these time-series enemy state presents multi-dimensional and massive characteristics. Using neural networks to learn enemy state information has become a research trend in the face of such traits. To address these challenges, we propose a hierarchical aggregation model to recognize the intention of the target. The bottom layer of our model is based on convolutional neural network(CNN) to perceive behavior features, and the middle layer is based on Bi-LSTM(Bi-directional long short-term memory) to aggregate the long-time interdependence information between sub-intentions. The top layer focuses on higher-level features that contribute more to the recognition of intent through the attention mechanism and finally combines the global information to recognize the intention. Extensive experimental results show the superiority of our model in that the recognition accuracy achieves 88.83%, which can solve the problem of identifying air target intent on the modern battlefield.

摘要

作战意图识别是指对敌方目标的状态信息进行分析, 从而解释和判断敌方想要达到的目的。随着作战平台信息化程度的不断提高, 这些具有时序性的敌方状态信息呈现多维、海量的特点。面对这样的特点, 提出基于神经网络的方法学习敌方状态信息。由于作战意图具有层次性, 并且意图行为之间具有依赖关系, 设计了一种层次聚合模型, 模型底层基于卷积神经网络感知行为特征, 中间层基于双向长短时记忆网络聚合子意图之间的长时依赖信息, 表达意图内部关系。顶层通过注意力机制将特征聚焦于对识别意图有更高贡献的高级特征, 最终感知全局信息以识别目标作战意图。实验数据表明, 相比其他网络结构, 提出的模型可以表达意图的层次性以及意图之间的长时依赖关系, 识别准确率可以达到88.83%, 适用于现代战场空中目标意图的识别问题。