针对在低信噪比(SNR)条件下,低截获概率雷达信号调制方式识别准确率低的问题,提出一种基于Transformer 和卷积神经网络(CNN)的雷达信号识别方法。首先,引入Swin Transformer 模型并在模型前端设计CNN 特征提取层构建了CNN+Swin Transformer 网络(CSTN),然后利用时频分析获取雷达信号的时频特征,对图像进行预处理后输入CSTN 模型进行训练,由网络的底部到顶部不断提取图像更丰富的语义信息,最后通过Softmax 分类器对六类不同调制方式信号进行分类识别。仿真实验表明:在SNR 为-18 dB 时,该方法对六类典型雷达信号的平均识别率达到了94.26%,证明了所提方法的可行性。

Aiming at the problem of low recognition accuracy of the low probability of intercept radar signal modulation method under the condition of low signal-to-noise ratio(SNR), a radar signal recognition method based on Transformer and convolutional neural network (CNN) is proposed. First, the Swin Transformer model is introduced and the CNN feature extraction layer is designed at the front end of the model to construct the CNN-Swin transformer network (CSTN). Then the time-frequency characteristics of radar signals are obtained by time-frequency analysis. The images are input into CSTN model for training after image preprocessing, and richer semantic information of images is continuously extracted from the bottom to the top of the network. Finally, six types of signals with different modulation modes are classified and recognized by Softmax classifier. Simulation experiments show that when the SNR is -18 dB, the average recognition rate of the method for six types of typical radar signals reaches 94. 26%, which proves the feasibility of the proposed method.

TN971

国防科技卓越青年科学基金资助项目(315090303)