【摘要】 针对智能电磁感知领域高质量训练数据稀缺的问题，提出一种基于残差注意力U型网络(U-Net)的条件生成对抗网络(RAU-cGAN)进行电磁信号样本重构的方法．该方法通过端到端学习从含噪时频图中恢复高保真信号，融合残差连接与自注意力机制以增强特征提取能力．实验结果表明，该方法在-20 dB低信噪比(SNR)环境下仍表现出稳健的重构性能，重构样本的峰值信噪比(PSNR)与结构相似性(SS)均达到高保真要求，为复杂电磁环境下的人工智能算法训练提供了高质量样本支撑．更多还原

【Abstract】 In order to solve the problem of scarcity of high-quality training data in the field of intelligent electromagnetic perception, this paper proposes a method using a conditional generative adversarial network based on a residual attention U-Net(RAU-cGAN) to carry out electromagnetic signal sample reconstruction. This method is used to recover high-fidelity signals from noisy time-frequency images through end-to-end learning, and integrate residual connection and self-attention mechanism to enhance feature extraction capability. Experimental results demonstrate that the proposed method still maintains robust reconstruction performance even in the environment of low signal-to-noise ratio(SNR) of-20 d B, and that the peak signal-to-noise ratio(PSNR) and structural similarity(SS) of the reconstructed samples both meet the high-fidelity requirements. And thus, this method provides high-quality sample support for the training of artificial intelligence algorithms in complex electromagnetic environments.更多还原