| 摘要: |
| 针对在复杂电磁环境下的雷达辐射源信号识别中传统方法失效，深度学习算法存在低信噪比下识别效果差、网络复杂参数多的问题，提出一种改进EfficientNet模型对雷达辐射源信号进行识别。首先引入卷积注意力模块(Convolutional Block Attention Module,CBAM)改进网络，提高网络对通道和空间特征的提取能力；结合h-Swish和ReLU两种激活函数进一步改进网络在保持网络精度的情况下加快网络训练；对样本进行标签平滑，对9种不同调制信号的时频图像做CutMix数据增强后输入网络，增强模型的泛化能力。通过实验分析，改进后的模型在信噪比为-10 dB的情况下对9种调制信号的综合识别率达到了94.24%，验证了该方法能够在低信噪比条件下有效识别雷达辐射源信号。 |
| 关键词: 雷达辐射源信号识别 改进EfficientNet 卷积注意力模块(CBAM) 标签平滑 |
| DOI：10.20079/j.issn.1001-893x.220321001 |
|
| 基金项目:国防科技卓越青年基金资助项目（315090303） |
|
| Radar signal modulation recognition based on improved EfficientNet |
| SU Congzhi,WANG Meiling,YANG Chengzhi,WU Hongchao |
| (School of Air Operations and Services,Aviation University of Air Force,Changchun 130022,China) |
| Abstract: |
| For the problem that traditional methods are failed in the recognition of radar emitter signal in complex electromagnetic environments and deep learning algorithm has poor recognition result under low signal-to-noise ratio(SNR) and too many parameters,an improved EfficientNet model is proposed to identify radar emitter signals.The Convolutional Block Attention Module(CBAM) is introduced to improve the original network to extract channel and spatial features,and the activation function of h-Swish and ReLU are used to further improve the network to speed up network training while maintaining network accuracy.Label smoothing is performed and the time-frequency images of 9 different modulation signals with CutMix data enhancement are used as the input,which enhances the generalization ability.Through experimental analysis,the improved model has a comprehensive recognition rate of 94.24% for the 9 modulation signals when the SNR is -10 dB,which verifies that the method can effectively identify the radar emitter signal under low SNR. |
| Key words: radar emitter signal recognition improved EfficientNet convolutional block attention module(CBAM) label smoothing |
