Underwater image real-time restoration model based on self-attention mechanism and GAN

Abstract

The invention relates to an underwater image real-time restoration model based on a self-attention mechanism and a GAN, and belongs to the technical field of deep learning. A generative adversarial network is used as a basic framework in the invention, the generative network adopts a coding-decoding structure, a synthetic image is generated at an original resolution through nine residual blocks and deconvolution operation, a self-attention module is introduced, richer advanced features can be captured to improve model performance, and in order to keep image content and remove underwater noise at the same time, the discrimination network adopts a structure of a multi-branch discriminator comprising an adversarial branch and a criticality branch. The self-attention module is added, the model structure is improved, the problem of low time efficiency of underwater image processing is solved ideally, and the method based on GAN-RS is trained to adapt to various underwater conditions, and has excellent real-time processing performance.

Description

technical field [0001] The invention relates to the technical fields of image processing and deep learning, and more specifically, the invention is oriented to the restoration task of underwater degraded images. Background technique [0002] In recent years, underwater robots have been widely used in marine resource exploration and other aspects. The enhancement and restoration technology of underwater degraded images has attracted much attention due to its significance for marine exploration, development and utilization. The complexity of the marine environment is diverse, and multiple unfavorable factors such as light transmission in water will be seriously attenuated by water absorption, reflection, and scattering. The collected underwater images will inevitably have limited visible range, blurred, Issues such as low contrast, non-uniform lighting, color incongruities, and noise. Underwater image enhancement and restoration technology aims to repair the degraded underwat...

AI Technical Summary

Problems solved by technology

The complexity of the marine environment is diverse, and multiple unfavorable factors such as light transmission in water will be seriously attenuated by water absorption, reflection, and scattering. The collected underwater images will inevitably have limited visible range, blurred, Problems such as low contrast, non-uniform lighting, color incongruity, and noise

Images