基于辅助分类器和变分自编码生成对抗网络的干扰识别

JAMMING RECOGNITION BASED ON AUXILIARY CLASSIFIER VARIATIONAL AUTO-ENCODING GENERATIVE ADVERSARIAL NETWORK

针对基于深度学习干扰识别方法在小样本集情况下性能恶化问题,提出一种基于辅助分类器和变分自编码生成对抗网络(AC-VAEGAN)的干扰识别方法。利用生成对抗网络和变分自编码器的核心思想设计识别模型,得到连续有意义的干扰样本集潜在空间;确定编码器、生成器和鉴别器的损失函数,且鉴别器采用动态学习率的优化算法,使得模型训练过程更加有效且稳定。仿真结果表明,在干扰时频图小样本数据集情况下,当干噪比为-10 dB~10 dB时,该方法对宽带噪声干扰、部分频带噪声干扰、单音干扰、多音干扰、脉冲干扰、跳频干扰、线性扫频干扰和二次扫频干扰这八种干扰的正确识别率均高于ACGAN和CNN。

To solve the performance deterioration of jamming recognition method based on deep learning in the case of small sample set,an interference recognition method based on auxiliary classifier variational auto-encoding generative adversarial network(AC-VAEGAN)is proposed.Using the core idea of generative adversarial network(GAN)and variational auto-encoder(VAE),AC-VAEGAN was designed to obtain latent space for continuous and meaningful jamming sample set.The loss function of encoder,generator and discriminator was modified,and the discriminator used an optimization algorithm of dynamic learning rate to make the training process of AC-VAEGAN more efficient and stable.The simulation results show that when the jamming-to-noise ratio is-10 dB~10 dB under the small sample dataset of jamming time-frequency diagrams,the correct recognition rate of the method to broadband noise jamming,partial band noise jamming,single-tone jamming,multi-tone jamming,pulse jamming,frequency hopping jamming,linear sweep jamming,quadratic sweep jamming is higher than that of ACGAN and CNN.