A method for compressing computational holograms using quantum neural networks with optimized initial weights

Abstract

The invention proposes a method for compressing and calculating a hologram by using a quantum neural network with optimized initial weight, and belongs to the technical field of compression and transmission of a calculation hologram. On the basis of compressing and transmitting computational holograms in the quantum BP neural network, the method uses the computational hologram training set to pre-train to obtain the optimized initial weight of the quantum BP neural network, and accelerates the convergence process of the pre-trained network by setting the pre-trained parameters to randomly initialize the variance. , and then use the optimized initial weights obtained by pre-training to perform secondary network fine-tuning training for the given holographic compressed data, and at the same time dynamically adjust the network learning rate during the optimization process to accelerate the quantum BP neural network compression transmission process. The present invention can use fewer iterations to complete the training of the compressed transmission network structure without changing the basic structure of the original quantum BP neural network, speed up the compression speed of the quantum BP neural network to calculate the hologram and ensure the reproduction of the hologram image the quality of.

Description

technical field [0001] The invention provides a method for compressing and computing a hologram using a quantum neural network with optimized initial weights, and specifically relates to the technical field of compression and transmission of a computing hologram. [0002] technical background [0003] The computational hologram method has the characteristics of flexibility, simplicity, and convenience. It avoids the complicated optical system and tedious preparation process of traditional optical holography, and can obtain effects that are difficult to achieve with artificially designed optical holography. The value of each point on the calculated hologram is the result of the interference between the diffracted wave and the reference light, covering all the information of the object, and each hologram contains a large amount of redundant information, which creates higher requirements for the storage and transmission of information , also limits the development of computation...

AI Technical Summary

Problems solved by technology

But the problem is that the quantum BP neural network is randomly initialized, which will make the initial weight of the network far from the optimal weight, and the network still needs more iterations to converge.

Images