Computational ghost imaging for atmospheric turbulence using physic mode-drive deep learning

Abstract

Atmospheric turbulence, a common natural phenomenon, causes beam drift due to its strong randomness and anisotropy, significantly hindering imaging applications. Computational ghost imaging (CGI) leverages the second-order correlation properties of light fields for indirect imaging, with low light source intensity requirements, single-pixel imaging, and excellent turbulence resistance. However, conventional CGI algorithms require multiple measurements to mitigate turbulence noise, resulting in a trade-off between acquisition number and image quality. To address this issue, we propose a physics-enhanced deep learning-based CGI method for atmospheric turbulence (PEACGI). In this approach, the ghost imaging algorithm is initially employed to generate a preliminary estimation that serves as input for the training network, allowing for a more comprehensive extraction of physical information. Additionally, we introduce a model-driven process to ensure that the reconstructed target closely approximates the real image. Theoretical and experimental results indicate that the proposed method exhibits robust performance across varying sampling ratios, turbulence intensities, and imaging distances. Consequently, our findings have significant implications for atmospheric turbulence imaging.