摘要
针对协同作战行动识别面临的行为特征空间大、模型参数多、训练速度慢等问题,将时空图模型和时序建模有机结合,提出基于深度时空循环神经网络的协同作战行动识别方法,建立战场协同作战行动识别架构,引入建议窗口生成机制划分战场空间为局部战场集,利用时空图设计层次循环神经网络模型以识别局部战场协同作战行动,并结合局部战场协同关系传递性实现整个战场的协同行动识别。实验分析表明,该方法具有较高的协同作战行动识别准确率。
To address the issues of large feature space, numerous model parameters and slow training speed in coordinated operation action recognition, a coordinated operational action recognition method based on a deep spatio-temporal recurrent neural network is proposed. In this method, a warped region generation mechanism is introduced to divide the whole battlefield into sub-battlefield. Meanwhile, a hierarchical recurrent neural network is constructed using spatio-temporal graph model, which is applied to the generated sub-battlefield to recognize coordinated operational action. Additionally, the recognized coordinated operational actions of sub-battlefields are merged to find out all coordinated operational actions based on the principle of transitivity of coordinated operational actions in local battlefield. Experiment results suggest that the proposed method possesses higher accuracy.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2018年第3期793-800,共8页
Journal of System Simulation
基金
国家自然科学基金(61773399
61374179
61703412)
军民共用重大研究计划联合基金(U1435218)
关键词
协同作战
层次循环神经网络
协同作战行为识别
时空图
态势理解
coordinated operation
hierarchical recurrent neural network
coordinated operation action recognition
spatio-temporal graph
situation comprehension
