An underwater image enhancement method based on a dual constraint generative adversarial network

By employing a dual-constraint generative adversarial network approach, this method utilizes datasets acquired by underwater AUVs and synthesized from underwater imaging models. By combining a generator, discriminator, and physical model, it addresses the issues of insufficient adaptability and poor image quality in existing underwater image enhancement methods, achieving underwater image enhancement with realistic colors and clear details.

CN117078544BActive Publication Date: 2026-07-21QINHUANGDAO PORT +1
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
QINHUANGDAO PORT
Filing Date
2023-08-18
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing underwater image enhancement methods can only adapt to a single underwater environment, and the generated images cannot simultaneously guarantee realistic colors and clear details.

Method used

A dual-constraint generative adversarial network is adopted. Real datasets are obtained through underwater AUVs and combined with underwater imaging models to synthesize datasets. The network is composed of two generators, a discriminator and a physical model, and a loss function is added for training. The generators are subject to dual constraints to cover different underwater scenarios.

Benefits of technology

It can recover images of different underwater scenes, and the generated enhancement results ensure both realistic colors and clear details. It covers most of the sea areas with synthetic underwater datasets and uses a background light prior method to solve the problem of inaccurate background light estimation in traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117078544B_ABST
    Figure CN117078544B_ABST
Patent Text Reader

Abstract

The application relates to an underwater image enhancement method based on a dual constraint generative adversarial network, which comprises the following steps: an underwater AUV is used to shoot underwater images on the seabed by matching a camera, so that the underwater images are obtained and a data set is formed; background light of different water area conditions is adopted to form a synthetic data set; two generators, two discriminators and a physical model are used to form a network; a loss function is added to the network model, and the generators are doubly constrained; the joint data set is used as the input of the network; the generators are trained, the parameters of the generators are fixed and unchanged, then the discriminators are trained, the discriminators are fixed, and then the generators are trained; after multiple iterations, the loss function tends to be minimum, the training is completed, corresponding weights are obtained, the network training is completed, and the output result of the input image or video is obtained. The application can solve the problem that the existing enhancement method can only adapt to a single underwater environment; the images of different underwater scenes can be restored, and the enhancement result can ensure that the color is lifelike and the details are clear.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Three-dimensional spectrum situation completion method and device based on generative adversarial network

    US20230386098A1