系统仿真学报 ›› 2022, Vol. 34 ›› Issue (7): 1459-1467.doi: 10.16182/j.issn1004731x.joss.21-0080

• 仿真建模理论与方法 • 上一篇    下一篇

基于深度学习的空中目标威胁评估方法

柴慧敏1,2(), 张勇2, 李欣粤1, 宋雅楠1   

  1. 1.西安电子科技大学计算机科学与技术学院,陕西 西安 710071
    2.光电信息控制和安全技术重点实验室,天津 300308
  • 收稿日期:2021-01-27 修回日期:2021-04-01 出版日期:2022-07-30 发布日期:2022-07-20
  • 作者简介:柴慧敏(1976-),女,博士,副教授,研究方向为信息融合与态势感知。E-mail:chaihm@mail.xidian.edu.cn
  • 基金资助:
    装备预研重点实验室基金(6142107190106)

Aerial Target Threat Assessment Method based on Deep Learning

Huimin Chai1,2(), Yong Zhang2, Xinyue Li1, Yanan Song1   

  1. 1.School of Computer Science and Technology, Xidian University, Xi'an 710071, China
    2.Science and Technology on Electro-Optical Information Security Control Laboratory, Tianjin 300308, China
  • Received:2021-01-27 Revised:2021-04-01 Online:2022-07-30 Published:2022-07-20

摘要:

针对空中目标威胁评估因素多、现有评估方法缺乏自学习能力的问题,采用深度学习理论建立了空中目标威胁评估的深层神经网络模型。为了提升模型训练的拟合效果,提出采用对称式的预训练方法,逐层地对模型中的隐含层进行预训练,最后对模型进行整体训练。分别通过样本测试集和空空仿真场景进行验证测试,结果表明:采用对称预训练方法,模型的威胁评估准确率高于其他三种预初始化方法;模型具有较好的鲁棒性,在无噪声下准确率大于90%,10%的正态噪声下,准确率大于70%。

关键词: 空中目标, 威胁评估, 深度学习, 对称式预训练

Abstract:

Due to many factors of aerial target threat assessment and the lack of self-learning ability of current assessment methods, a deep neural network model for aerial target threat assessment is established using deep learning theory. In order to improve the fitting effect of the model training, a symmetric pre-training method is given. The hidden layers of the model are pre-trained layer by layer, and finally the whole model is trained. Sample data and air to air simulation scene experiments are carried out respectively. The experiments results show that the accuracy of the model using the symmetric pre-training method is higher than the other three initialization methods. The accuracy of the model is more than 90% without noise and more than 70% under 10% normal noise, which shows its better robustness.

Key words: aerial target, threat assessment, deep learning, symmetrical pre-training

中图分类号: