摘要
在雷达辐射源信号识别中,针对现有的识别方法存在实时性差、网络模型参数量大以及难以应用于资源受限的设备等问题,提出了一种基于轻量级卷积神经网络的雷达辐射源信号识别方法。首先,利用平滑伪Wigner-Ville分布(smooth pseudo Wigner-Ville distribution,SPWVD)将雷达辐射源信号转换为时频图像,并对时频图像进行图像预处理;其次,基于Vision Transformer的架构设计,结合传统的卷积神经网络,构建了轻量级网络模型RecNet;最后,利用预处理后的时频图像对RecNet网络模型进行训练,实现对9种雷达辐射源信号的高效识别。实验表明,该方法在信噪比为−8 dB时,对9种雷达辐射源信号的识别准确率达到95.7%,模型参数量为0.9×10^(6)且推理延迟仅为4.67 ms,在保证较高识别准确率的同时,具有更快的识别速度和更小的模型参数量,具有一定的工程应用价值。
In the recognition of radar radiation source signals,the existing recognition methods have problems such as poor real-time performance,large number of network model parameters,and difficulty in applying to resource-constrained devices.This paper proposes a radar radiation source signal recognition method based on a lightweight convolutional neural network—Recognition Network(RecNet).First,the radar radiation source signal is converted into a time-frequency image using a smooth pseudo Wigner-Ville distribution(SPWVD),and the time-frequency image is preprocessed.Secondly,based on the architecture design of Vision Transformer,and combined with the traditional convolutional neural network,a lightweight network model RecNet is constructed.Finally,the preprocessed time-frequency image is used to train the RecNet network model,achieving high-efficiency recognition of radar radiation source signals.Experiments show that when the signal-to-noise ratio is−8 dB,the recognition accuracy of 9 types of radar radiation source signals reaches 95.7%,the model parameter volume is 0.9 MB,and the inference delay is only 4.67 ms.While ensuring a high recognition accuracy,this method has a faster recognition speed and a smaller model parameter volume,and has certain engineering application value.
作者
张忠民
姜嵛涵
ZHANG Zhongmin;JIANG Yuhan(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)
出处
《应用科技》
2025年第1期166-172,共7页
Applied Science and Technology
