|本期目录/Table of Contents|

[1]胡奕彬,胡樾.融合小波变换与多尺度残差的水下图像增强[J].集美大学学报(自然科学版),2026,31(5):605-619.
　HU Yibin,HU Yue.Underwater Image Enhancement via Wavelet Transform and Multi-Scale Residuals[J].Journal of Jimei University,2026,31(5):605-619.
点击复制

融合小波变换与多尺度残差的水下图像增强(PDF)
分享到：

《集美大学学报（自然科学版）》[ISSN:1007-7405/CN:35-1186/N]

卷:
第31卷
期数:
2026年第5期
页码:
605-619
栏目:
海洋信息与数理科学
出版日期:
2026-09-28

文章信息/Info

Title:
Underwater Image Enhancement via Wavelet Transform and Multi-Scale Residuals
作者:
胡奕彬; 胡樾
（集美大学诚毅学院,福建 厦门 361021）
Author(s):
HU Yibin; HU Yue
（Chengyi College，Jimei University,Xiamen 361021,China）
关键词:
水下图像增强; 小波变换; 注意力机制; 颜色分布; 多尺度残差网络
Keywords:
underwater image enhancement;  wavelet transform;  attention mechanism;  color distribution;  multi-scale residual network
分类号:
-
DOI:
-
文献标志码:
A
摘要:
水下图像因水体对光的选择性吸收与悬浮粒子的散射作用，普遍存在色偏、对比度低、细节模糊等退化问题，严重影响视觉质量与下游视觉任务性能。本文提出一种融合小波变换与多尺度残差的水下图像增强方法。该方法将增强任务迁移至小波系数域，利用离散小波变换分离图像低频结构与高频细节；构建以小波系数为输入的改进UNet网络，嵌入多尺度残差增强模块与通道注意力机制，实现多尺度退化特征的有效提取与融合；设计基于巴氏距离的颜色分布损失函数，保证生成图像与参考图像的整体色调一致性。在EUVP和UIEB数据集上的实验表明，本文方法在PSNR、SSIM、UIQM等客观指标上取得了具有竞争力的结果，特征匹配及主观视觉对比进一步验证了其在色彩还原、对比度提升和细节恢复方面的有效性。
Abstract:
Underwater images commonly suffer from color distortion,low contrast,and blurred details due to the selective absorption of light by water bodies and the scattering effects of suspended particles,which significantly degrade visual quality and downstream vision task performance.This paper proposes an underwater image enhancement method that integrates wavelet transform and multi-scale residual learning.The enhancement task is migrated to the wavelet coefficient domain,where discrete wavelet transform separates low-frequency structures from high-frequency details.An improved UNet architecture taking wavelet coefficients as input is constructed,embedding multi-scale residual enhancement modules and channel attention mechanisms for effective extraction and fusion of multi-scale degradation features.A color distribution loss function based on Bhattacharyya distance is designed to constrain the overall color consistency between the generated and reference images.Experiments on the EUVP and UIEB datasets demonstrate that the proposed method achieves competitive performance on PSNR,SSIM,and UIQM metrics.Feature matching and subjective visual comparisons further validate its effectiveness in color restoration,contrast enhancement,and detail recovery.

参考文献/References:

相似文献/References:

[1]赵伟,陈仁安,张晓玲,等.基于ICA和DWT的自适应盲水印算法[J].集美大学学报(自然科学版),2012,17(4):310.
　ZHAO Wei,CHEN Ren-an,ZHANG Xiao-ling,et al.Adaptive Blind Watermarking Algorithm Based on ICA and DWT[J].Journal of Jimei University,2012,17(5):310.
[2]浦云明,汪志华,乔峰,等.ENS3000噪声监测系统研究与实现[J].集美大学学报(自然科学版),2015,20(3):236.
　PU Yun-ming,WANG Zhi-hua,QIAO Feng,et al.The Research and Implementation of ENS3000 Noise Monitoring System[J].Journal of Jimei University,2015,20(5):236.
[3]邱秀亮,肖世校,张保灿.基于小波神经网络的手足口病发病率预测研究[J].集美大学学报(自然科学版),2017,22(3):69.
　QIU Xiuliang,XIAO Shixiao,ZHANG Baocan.Predicting the Incidence of Hand-foot-mouth Disease Based on Wavelet Neural Network[J].Journal of Jimei University,2017,22(5):69.

备注/Memo

备注/Memo:
更新日期/Last Update: 2026-11-01