The invention provides an amplitude hologram generation method based on
deep learning and camera
closed loop. The method comprises the following steps: firstly, acquiring
light field intensity information of a to-be-reconstructed target, and generating an initial amplitude type hologram according to modulation characteristics of a
spatial light modulator; the amplitude hologram is loaded to a modulator, reconstruction is achieved through
optical diffraction, and a reconstructed image is collected in real time through a camera; constructing a
loss function based on the deviation between the reconstructed image and the target image, performing camera closed-
loop optimization on the hologram through an end-to-end gradient backhaul mechanism, and realizing the derivability of the binarization process by adopting a micro-
approximation function under the condition that the
binary modulation is not derivable; and training a
deep learning network by using a propagation model obtained by closed-
loop optimization, so that the network can quickly predict a high-quality amplitude hologram. The method compensates non-ideal deviation of an optical
system while suppressing
speckle noise and conjugate image interference, is suitable for
holographic display of monochromatic and colorful and two-dimensional and three-dimensional scenes, and has the advantages of high reconstruction quality, efficient optimization, strong adaptability and the like.