Metasurfaces are widely applied in various applications, such as none‐line‐of‐sight detection, radar imaging enhancement, and non‐invasive monitoring. However, electromagnetic (EM) information recovery in inaccessible and occluded areas is of great importance to obtain complete EM picture, albeit challenging. Conventional methods to this end typically necessitate specific prior knowledge and suffer from performance degradation due to implicit computation mechanism. Here a flow‐based framework is proposed to facilitate the explicit computation of conditional distribution between the partially accessible EM field and complete EM field. The adjacent distributions in a hierarchical architecture exhibit similarity and seamless convertibility between each other, facilitating a smooth transition without performance degradation. The method is benchmarked through two typical scenarios, i.e., resolution enhancement and field recovery in randomly occluded areas. Even in an entirely unseen scene, the EM information recovery maintains consistence with the ground truth, with maximum error below 10%. The work provides a key advance for EM information recovery in complex real‐world environment, offering fresh insights on information access and detection even in extreme cases.

中文翻译：

---

基于流的电磁信息恢复，适用于无法进入的区域和低分辨率检测


超表面广泛应用于各种应用，如非视距检测、雷达成像增强和无创监测。然而，尽管具有挑战性，但在无法接近和遮挡的区域恢复电磁 （EM） 信息对于获得完整的 EM 图像非常重要。为此，传统方法通常需要特定的先验知识，并且由于隐式计算机制而遭受性能下降。这里提出了一个基于流的框架，以促进部分可访问的 EM 场和完整的 EM 场之间的条件分布的显式计算。分层架构中的相邻分布表现出相似性和彼此之间的无缝可转换性，有助于平稳过渡而不会降低性能。该方法通过两个典型场景进行基准测试，即随机遮挡区域的分辨率增强和场恢复。即使在完全看不见的场景中，EM 信息恢复也与地面实况保持一致，最大误差低于 10%。这项工作为复杂现实世界环境中的 EM 信息恢复提供了关键进展，即使在极端情况下也能为信息访问和检测提供新的见解。