Research on Computer Generated Hologram Based on Deep Learning

Compressed light and entangled light are very important resources in quantum optics, and have attracted much attention due to their application prospects in many fields. Among the many methods of generating holograms, computer generated hologram technology has attracted much attention, and the optimization algorithm of computer generated hologram is the research hotspot nowadays. In recent years, with the rise of deep learning technology, more and more researchers began to apply deep learning to computer generated hologram to optimize the algorithm of computer generated hologram. However, the current models applied to computer-generated holograms are almost all based on convolutional neural networks, and there are many excellent deep learning models worth trying. In this paper, a computer-generated hologram model based on Vision Transformer is proposed by combining optical shaping system with computer-generated hologram technology and using deep learning technology for optimization. In this model, the physical process of the beam shaping system is innovatively added, so that the computer-generated hologram is combined with the hardware. The model in this paper is trained using Mnist handwritten digital data set, and then the trained model is used to predict the test set to generate hologram, and then the hologram is used to reconstruct the target light field. The reconstructed images verify the feasibility of the model in generating holograms to achieve target light field reconstruction. The SSIM index is used to evaluate the quality of reconstructed images. The results show that the average SSIM of reconstructed holograms corresponding to all images in the test set is 0.751. The relevant work in this paper enriches the research of deep learning applied to computer-generated holograms, provides a new solution for the preparation of higher-order modes, and also provides a new direction for the generation of higher-quality holograms in the future.