基于深度残差神经网络的无线电引信多调制类型时域混叠信号识别方法  

A Time Domain Overlapping Signal Recognition Method for Multiple Modulation Types Radio Fuze Based on Deep Residual Neural Network

作　　者：常仁 朱玉鹏 周辉 刘金生 CHANG Ren;ZHU Yupeng;ZHOU Hui;LIU Jinsheng(Academy of Military Sciences Institute of Systems Engineering,Beijing 100097,China)

机构地区：[1]中国人民解放军军事科学院系统工程研究院,北京100097

出　　处：《探测与控制学报》2024年第6期37-45,共9页Journal of Detection & Control

基　　金：国防预先研究项目(315095201)。

摘　　要：针对多方向不同弹药来袭场景下产生的多调制类型时域混叠无线电引信信号识别问题,提出一种基于深度残差神经网络的时域混叠引信信号自动识别方法,实现低信噪比下多调制类型引信时域混叠信号的精确识别。采用DnCNN编码解码结构对时频域混沌信号进行降噪,为低信噪比下多标签信号有效识别奠定基础;对于可匹配目标信号,建立多调制类型引信时域混叠信号多标签分类模型,构建基于深度残差神经网络的引信多调制类型时域混叠信号识别模型;对于不匹配目标信号,建立增量式小样本学习方法,在不影响原有模型参数条件下,通过增加额外的增量学习结构,实现对新出现的不匹配引信信号的增量学习与在线识别。仿真结果表明,该方法能够在低信噪比下实现不同调制类型引信时域混叠信号的精确识别,-10 dB信噪比下平均识别率可达90%。A method based on deep residual neural network for automatic modulation recognition of multi modulation type time-domain aliasing radio fuze signals in scenarios of multiple incoming ammunition from different directions was proposed to achieve accurate recognition of multi modulation type time-domain aliasing signals under low signal-to-noise ratio.A denoising preprocessing method for aliased signals based on feedforward denoising convolutional neural network(DNCNN)encoding decoding was proposed,laying the foundation for effective signal recognition under low signal-to-noise ratios;A multi modulation type fuze time-domain aliasing signal multi label classification model was established for matching target signals,and deep semantic feature extraction was performed on the denoised time-frequency images to achieve automatic classification and recognition of time-domain aliasing signals;An incremental small sample learning method was proposed for mismatched target signals.By adding an additional incremental learning structure without affecting the original model parameters,incremental learning and online recognition of new mismatched fuze signals are achieved.The simulation results showed that this method could achieve accurate recognition of time-domain aliasing signals of different modulation types at low SNR,and the average recognition rate could reach 90%at-10 dB SNR.

关 键 词：无线电引信 多调制类型时域混叠信号 深度残差神经网络 不匹配目标信号 

分 类 号：TJ434.1[兵器科学与技术—火炮、自动武器与弹药工程]