Image reconstruction of scattered vortex light field based on deep learning

Vortex beams propagating through a scattering medium generate a random field that obstructs the retrieval of original information carried by the vortex beam. In this paper, a new method based on an improved generative adversarial network (GAN) to reconstruct scattered object information carried by a vortex beam is proposed. First, particular holograms were designed to achieve imaging while preserving the vortex beam's orbital angular momentum property. Then, an improved GAN was used to realize the reconstruction from speckle images after the vortex beams pass through the scattering medium. In order to evaluate the image reconstruction ability of the proposed GAN, three types of vortex beams with different topological charges and three types of scattering media with different roughness are employed in experiments, and 15,000 speckle-object pairs collected are used to train five GAN models. The results show that each test dataset's average Pearson correlation coefficient reaches 0.886, 0.934, 0.922, 0.903, and 0.886, which means that the formed speckle field of vortex beam still retains certain information and the improved GAN have excellent ability in restoring original information from scattered vortex light field.