Ghost imaging through multiple scattering media at low sampling rates via an enhanced UNet

To address the challenges of imaging through scattering media, we propose GI_UNet, a differential ghost imaging (DGI) framework that integrates Monte Carlo photon-transport modeling with an enhanced UNet. By utilizing high-fidelity point spread functions (PSFs), the framework predicts the measurement matrix of trainable speckles after their propagation through scattering media, replacing the ideal illumination patterns used in traditional DGI. Subsequently, the enhanced UNet suppresses scattering noise and restores edge details through its optimized architectural components. Simulations and experiments demonstrate that GI_UNet outperforms existing mainstream methods. Given its imaging robustness under scattering conditions and at low sampling rates, GI_UNet provides an effective solution for various scattering-limited imaging scenarios.