Phase‐Modulated Terahertz Compressive Imaging With Physics‐Integrated Deep Unfolding

ABSTRACT Terahertz (THz) compressive imaging is fundamentally limited by the absence of practical spatial light modulators capable of implementing native ±1 phase‐mask encoding. Consequently, most existing systems rely on amplitude‐only modulation and complementary measurements to synthesize ±1 masks, thereby doubling acquisition time and reducing measurement efficiency. Here, we present a THz compressive imaging framework based on a liquid‐crystal programmable metasurface that enables direct binary phase modulation for native signed‐mask encoding. This phase‐domain operation eliminates the need for complementary measurements and significantly improves acquisition efficiency. To address the sign ambiguity inherent in amplitude‐only detection, a physics‐guided deep neural network is introduced for robust image reconstruction. Both simulations and experiments demonstrate high‐fidelity imaging at a sampling ratio as low as 37.5%, highlighting a practical route toward efficient and scalable THz computational imaging.