基于卷积神经网络的双阶段水下图像增强方法
Two-stage underwater image enhancement method based on convolutional neural networks
路斯棋 1管凤旭 1赖海涛 1杜雪2
作者信息
- 1. 哈尔滨工程大学智能科学与工程学院,哈尔滨 150001
- 2. 哈尔滨工程大学智能科学与工程学院,哈尔滨 150001;哈尔滨工程大学教育部极端海洋环境波动场前沿科学中心,哈尔滨 150001
- 折叠
摘要
由于水体对光线不同粒子的吸收能力具有一定差异,水下采集到的图像往往存在严重的退化现象,严重影响水下机器人对环境的感知.传统的图像处理方法和基于退化模型的图像恢复算法受到水下环境的复杂性和物理参数不确定性的影响往往表现出较差的泛化能力.为提高水下图像的视觉效果,利用深度学习模型强大的学习能力,提出一种基于卷积神经网络的双阶段水下图像增强方法,通过图像损坏和图像恢复两个阶段的处理将退化的水下图像增强为视觉效果优秀的近空气图像.在Challenge60、U45、EUVP和RUIE数据集上的测试结果表明,提出的方法相比于已有水下图像复原、增强算法具有更好的增强效果,水下图像质量指标(UIQM)提升了 5.18%,水下彩色图像质量评价(UCIQE)指标提升了6.64%.
Abstract
Images taken underwater frequently suffer from substantial degradation due to the varied capabilities of water particles to absorb light,which has a significant impact on how underwater robots perceive their surroundings.The intricacy of underwater environments and uncertainties in physical factors usually result in poor generalization for traditional image processing techniques and degradation model-based picture restoration systems.A two-stage underwater image enhancement technique based on convolutional neural networks(CNNs)is suggested to improve the quality of underwater images.This method improves degraded underwater images into visually superior near-air images through damage and restoration phases.Testing results on Challenge60,U45,EUVP,and RUIE datasets show that the proposed method achieves better enhancement compared to existing underwater image restoration and enhancement algorithms,with improvements of 5.18%and 6.64%respectively for UIQM and UCIQE scores.
关键词
卷积神经网络/深度学习/双阶段/水下图像增强/水下图像复原Key words
convolutional neural networks/deep learning/two-stage/underwater image enhancement/underwater image restoration引用本文复制引用
出版年
2025
国家科技期刊平台
NSTL
维普
万方数据