首页|期刊导航|电子科技|基于颜色校正的GAN水下图像增强网络

基于颜色校正的GAN水下图像增强网络OA

Underwater Image Enhancement Algorithm Based on GAN-Fused Color Correction

中文摘要英文摘要

针对光的散射和吸收导致水下成像技术存在颜色失真与细节模糊等问题,文中提出了一种基于GAN(Gen-erative Adversarial Network)改进的水下图像增强网络(Underwater Image Color Correction Enhancement Network,UICEN-GAN).不同于传统GAN,所提网络在生成器首部添加宽范围感知模块和双重注意力机制,宽范围感知模块采用较大的卷积核覆盖更大的接受域,双重注意力机制通过将空间、通道注意力并联来增强图像的细节信息.尾部添加颜色校正模块,该模块采用自注意力计算分配相关权值,将图像特征映射到RGB颜色空间,从而解决颜色失真问题.实验结果表明,所提网络的增强结果优于其他先进网络,增强后图像的峰值信噪比、结构相似性指数相比指标次之的网络分别提升了10.6%和1.8%,在主观评价和客观评价方面均取得了良好的效果.

In view of the problems such as color distortion and detail blurring caused by light scattering and absorption in underwater imaging technology,an improved UICEN-GAN(Underwater Image Color Correction En-hancement Network)based on GAN(Generative Adversarial Network)is proposed.Unlike traditional GAN,the proposed network adds a wide-range perception module and a dual attention mechanism at the generator header.The wide-range perception module uses a larger convolutional kernel to cover a larger receiving domain,and the du-al attention mechanism enhances the detailed information of the image by paralleling spatial and channel attention.A color correction module is added at the end.This module uses self-attention calculation to allocate relevant weights and maps image features to the RGB color space,thereby solving the problem of color distortion.The experi-mental results show that the enhancement effect of the proposed network is superior to that of other advanced net-works.The peak signal-to-noise ratio and structural similarity index of the enhanced image have increased by 10.6%and 1.8%respectively when compared with the network with the second-highest index.Good effects have been achieved in both subjective and objective evaluations.

亢淑洁;陈清江

西安建筑科技大学 理学院,陕西 西安 710055西安建筑科技大学 理学院,陕西 西安 710055

信息技术与安全科学

水下图像增强生成对抗网络颜色校正注意力机制卷积神经网络深度学习图像处理端到端

underwater image enhancementGANcolor correctionattention mechanismconvolutional neural networkdeep learningimage processingend-to-end

《电子科技》 2026 (10)

42-51,10

国家自然科学基金(12101482) National Natural Science Foundation of China(12101482)

10.16180/j.cnki.issn1007-7820.2026.10.005

评论