航空学报 > 2023, Vol. 44 Issue (24): 328551-328551   doi: 10.7527/S1000-6893.2023.28551

代价敏感的空中目标意图识别方法

丁鹏, 宋亚飞()   

  1. 空军工程大学 防空反导学院,西安 710051
  • 收稿日期:2023-02-13 修回日期:2023-02-27 接受日期:2023-04-17 出版日期:2023-04-23 发布日期:2023-04-21
  • 通讯作者: 宋亚飞 E-mail:yafei_song@163.com
  • 基金资助:
    国家自然科学基金(61806219);陕西省科学基金(2021JM-226);陕西省高校科协青年人才托举计划(20190108);陕西省创新能力支撑计划(2020KJXX-065)

A cost-sensitive method for aerial target intention recognition

Peng DING, Yafei SONG()   

  1. Air and Missile Defense College,Air Force Engineering University,Xi’an 710051,China
  • Received:2023-02-13 Revised:2023-02-27 Accepted:2023-04-17 Online:2023-04-23 Published:2023-04-21
  • Contact: Yafei SONG E-mail:yafei_song@163.com
  • Supported by:
    National Natural Science Foundation of China(61806219);National Science Foundation of Shaanxi Provence(2021JM-226);Young Talent Fund of University and Association for Science and Technology in Shaanxi(20190108);Innovation Capability Support Plan in Shaanxi(2020KJXX-065)

摘要:

空中意图识别是防空作战指挥决策的基本依据,也是战场认知和智能决策的前提和基础。然而,现有的研究大单一地追求意图识别的准确率,没有考虑意图识别的误判代价是否等价。针对空中目标作战意图识别在现代防空作战决策中的重要作用和现阶段意图识别面临的没有考虑误判代价的问题,设计了一种针对空中目标时序数据的GRU-FCN模型提高识别能力,同时提出了一种针对代价敏感意图识别模型的CAIR改进方法。首先,通过分析防空作战的特点,获取空中目标的意图特征并进行预处理。其次,利用GRU模型捕捉时序特征,FCN模型提取数据中的复杂特征,加入CAIR算法使模型具备代价敏感意图识别的能力。最后,通过仿真实验、消融实验、对比实验验证了本文所提模型的效果。实验分析表明:GRU-FCN模型的准确率达到了98.57%,超过对比实验中其他空中目标意图识别模型2.24%;融入代价敏感改进策略后,整体误判代价由0.346 7降为0.175 7,在保证了准确率满足要求的同时具备了代价敏感意图识别的能力。

关键词: 深度学习, 代价敏感, 空中目标, 意图识别, 门控循环单元(GRU), 全卷积网络(FCN)

Abstract:

Air intention recognition is the key to the transition from the information domain to the cognitive domain, serving as the basis for command and decision making in air defense operations, as well as the prerequisite and foundation for battlefield cognition and intelligent decision making. It has always been considered the core content of battlefield situational awareness. However, existing research mostly focuses solely on the accuracy of intention recognition, without considering whether the misjudgment costs of intention recognition are equivalent. Air target intention recognition should consider both accuracy and cost sensitivity. To solve the problem of different misjudgments that may cause different losses to our side, a new deep learning model (GRU-FCN) for cost sensitive aerial target intention recognition is designed based on gated cyclic units and fully convolutional networks, and a cost sensitive improvement strategy is also proposed. Experimental analysis shows that the accuracy of the GRU-FCN model reaches 98.57%, surpassing other aerial target intention recognition models in the comparative experiment by 2.24%. After incorporating the cost sensitive improvement strategy, the overall misjudgment cost has been reduced from 0.346 7 to 0.175 7, ensuring that the accuracy meets the requirements while also possessing the ability to recognize cost sensitive intentions.

Key words: deep learning, cost sensitive, aerial target, intention recognition, Gated Recurrent Unit (GRU), Fully Convolutional Networks (FCN)

中图分类号: