Synthetic aperture imaging systems can be applied to equivalently get high-resolution images of traditional monolithic primary mirror systems with less weight and costs. Spatial modulation diversity (SMD), a newly developed post-processing technology applicable for such synthetic aperture systems, is very sensitive to Gaussian noise, which greatly limits its further application. In this paper, we propose an improved SMD strategy by introducing the deep denoising convolutional neural networks (DnCNN) into the image preprocessing to improve the robustness of SMD. Results of the numerical simulations demonstrate that the strategy proposed exhibits superior performance compared to the traditional SMD technique in terms of both the root-mean-square error (RMSE) of phase estimates and the structural similarity (SSIM) of image reconstruction. In view of the superiority and robustness, the method we proposed may have important application prospects in multi-aperture imaging systems.

中文翻译：

---

基于深度去噪卷积神经网络的高噪声鲁棒空间调制分集算法

合成孔径成像系统可以应用于以较少的重量和成本等效地获得传统的单片主镜系统的高分辨率图像。空间调制分集（SMD）是一种适用于此类合成孔径系统的最新开发的后处理技术，对高斯噪声非常敏感，这极大地限制了其进一步的应用。在本文中，我们通过将深度降噪卷积神经网络（DnCNN）引入图像预处理中以提高SMD的鲁棒性，提出了一种改进的SMD策略。数值模拟结果表明，与传统的SMD技术相比，所提出的策略在相位估计的均方根误差（RMSE）和图像重建的结构相似性（SSIM）方面均表现出优异的性能。