Electromagnetic Perception With Artificial Intelligence‐Empowered Programable Metasurface: From Data‐Driven Imaging to Semantic Sensing

Electromagnetic (EM) perception with artificial intelligence (AI)‐empowered programable metasurfaces has become a pivotal technique in the modern society, owing to their capability of intelligent EM manipulation in response to the changing environment and user's demand. Specifically, the self‐adaptivity and computational compatibility of AI‐empowered programable metasurfaces enable the physical‐level adaptive data acquisition and digital‐level postprocessing, which can be deeply integrated with data‐driven methods to achieve novel perception systems that outperform perception systems based on traditional inversion methods in challenging scenarios. More recently, the integration with semantic priors has emerged as a new trend in intelligent EM perception, leading to the higher‐level cognitive capabilities. This review presents the evolution of intelligent EM perception systems through the fusion of AI‐empowered programable metasurfaces with progressively sophisticated machine learning—from implicit data priors to explicit semantic priors—and systematically reviews key advances in this progression.