基于监督对比学习的无线电引信干扰识别方法  

A recognition method of radio fuze signal based on supervised contrastive learning

作　　者：钱鹏飞 秦高林[3] 陈齐乐 郝新红[1,2] QIAN Pengfei;QIN Gaolin;CHEN Qile;HAO Xinhong(Science and Technology on Electromechanical Dynamic Control Laboratory,Beijing Institute of Technology,Beijing 100081,China;Tangshan Research Institute,Beijing Institute of Technology,Tangshan 063099,China;North China University of Water Resources and Electric Power,Zhengzhou 450045,China)

机构地区：[1]北京理工大学机电动态控制重点实验室,北京100081 [2]北京理工大学唐山研究院,唐山063099 [3]华北水利水电大学,郑州450045

出　　处：《北京航空航天大学学报》2025年第3期953-961,共9页Journal of Beijing University of Aeronautics and Astronautics

摘　　要：连续波调频多普勒引信在战场上容易受到干扰,从而导致弹药早炸失去毁伤能力。为了提高调频多普勒引信对信息型干扰的抗干扰能力,实现多种干扰信号与目标回波的区分,提出一种基于监督对比学习的目标与干扰信号分类识别方法。该方法首先通过残差网络和自注意力机制搭建了主干网络;然后利用引入标签的方式改进了对比学习损失函数,实现了监督对比学习;最后采用中频信号搭建数据集,通过监督对比学习的方式来训练网络,从而实现对目标与干扰信号的分类和识别。仿真结果表明:该方法能够实现多种干扰种类与目标回波的识别,并且识别率能够达到98.7%。在低信噪比环境下的识别效果更为出色,在信噪比为−18 dB的环境下,仍然能有91.81%的识别率,相比普通残差网络的86.12%的识别率更高。Frequency modulated continuous wave(FMCW)Doppler fuze is easy to be interfered with on the battlefield,resulting in an early explosion and loss of damage ability.To improve the anti-jamming ability of FMCW Doppler fuze against information-based jamming and realize the distinction between multiple jamming signals and target echoes,this paper proposed a method of target and jamming signal classification and recognition based on supervised contrastive learning.Firstly,the backbone network was constructed by residual network and self-attention mechanism.Then,the contrastive learning loss function was improved by introducing labels,and supervised contrastive learning was realized.Finally,an intermediate frequency signal was used to build the dataset,and the network was trained by supervised comparative learning,so as to realize the classification and recognition of the target and jamming signal.The simulation results show that this method can realize the recognition of multiple jamming types and target echoes,and the recognition rate can reach 98.7%.In the low signal-to-noise ratio(SNR)environment,the recognition effect is better.In the SNR environment of−18 dB,the recognition rate is still 91.81%,which is higher than the 86.12%recognition rate of ordinary residual networks.

关 键 词：调频多普勒引信 电子对抗 深度神经网络 监督对比学习 信号识别 

分 类 号：TJ434.1[兵器科学与技术—火炮、自动武器与弹药工程]