基于多级跳线残差网络的雷达辐射源识别

针对复杂体制雷达辐射源的识别问题, 提出了一种基于时频特征提取与多级跳线残差网络(multi-level jumper residual network, MLJ-RN)结合的识别方法。首先，计算辐射源信号的平滑伪Wigner-Ville时频分布生成时频图像以表达信号本质特征, 将图像进行预处理以保留信号细微特征差异。然后，设计多级跳线连接的残差单元, 在此基础上构造MLJ-RN, 对时频图像相邻卷积层的细微特征进行学习和识别, 并使用随机梯度下降法训练网络。最后，通过对网络进行参数优化, 强化对信号的深层特征提取能力。仿真结果表明, 信噪比为－5 dB时, 该方法对12类雷达辐射源信号的整体识别概率达到95.1%, 从而验证了该方法在低信噪比下识别雷达信号的有效性。

Radar emitter recognition based on multi-level jumper residual network

Aiming at the problem of complex radar emitter recognition, a recognition method based on time-frequency feature extraction and multi-level jumper residual network (MLJ-RN) is proposed. Firstly, the smoothed pseudo Wigner-Ville distribution (SPWVD) of emitter signals is calculated, a time-frequency image is generated to express the essential characteristics of the signal, and the image is preprocessed to retain the subtle differences of the signal features. Then a residual unit connected by multi-level jumpers is designed, a multi-level jumper residual network is constructed on this basis, the fine features of the adjacent convolution layers of the time-frequency image are learned and recognized, and the random gradient descent method is used to train the residual network. Finally, through the adjustment of the network, the parameters of the network are optimized to enhance the deep feature extraction ability of the signal. The simulation results show that when the signal-to-noise ratio (SNR) is －5 dB, the overall recognition probability of the method for 12 types of radar emitter signals reaches 95.1%, which verifies the effectiveness of the method in identifying radar signals at low SNRs.