Unlocking Robust Structured Light in Complex Media Through Personalized Search With Physics‐Informed Neural Networks

ABSTRACT

Structured light undergoes severe distortion when propagating through complex media such as turbulent atmospheres and biological tissues, which hinders its practical deployment. Here, we propose a tailored search paradigm based on spatial‐mode‐fitted physics‐informed neural network (PINN) to retrieve structured light fields robust to such distortions in designated complex media. Different from conventional pixel‐based methods, our approach represents the optical field as a superposition of orthogonal Laguerre–Gaussian modes, which substantially reduces dimensionality and computational cost. The proposed physics‐driven and label‐free PINN incorporates wave propagation laws into the loss function, enabling the identification of light fields whose mode compositions remain invariant after transmission through disordered media. Experimental results demonstrate that the obtained robust light fields achieve an intensity structural similarity index measure (SSIM) above 0.8 and a negative Pearson correlation coefficient (NPCC) of the mode power spectrum below 0.1 in turbulent media, a performance nearly comparable to that in free‐space propagation. By customizing the loss function, this framework can also generate light fields that meet practical requirements such as a reduced number of constituent modes. This efficient and versatile strategy opens up new avenues for the deployment of robust structured light in real‐world scattering environments.