Higher-Order Spatial Mode Detection Leveraging Deep Learning on Random Optical Patterns

Laguerre-Gauss (LG) beams, characterized by their helical wavefronts and carrying orbital angular momentum (OAM) of ℓℏ where ℓ is the topological charge (TC), hold significant promise for optical communication, imaging, and quantum information science. However, accurately detecting higher topological charges (ℓ ≤ 50) when distorted by scattering media remains a substantial challenge due to wavefront distortion and speckle formation. This work addresses this limitation by proposing and evaluating two deep learning architectures, a Convolutional Neural Network (CNN) based model and a Vision Transformer (ViT) based model. Experimental results show that the ViT based model achieves a better mean classification accuracy of 98.1%, outperforming the CNN based model by approximately 3.3%.Additionally, we validated the robustness of the proposed models by accurately detecting the sign of TC along with the magnitude. Unlike existing approaches, our method detects the TC using only a patch of the distorted intensity profile rather than the full beam. This capability, especially for high TCs in scattering environments, shows the way for more reliable, high capacity optical systems in real world applications.