基于CycleGAN的无监督图像去雾网络

doi:10.16411/j.cnki.issn1006-7736.2025.03.008

PDF(25006 KB)

大连海事大学学报 ›› 2025, Vol. 51 ›› Issue (3) : 74-84. DOI: 10.16411/j.cnki.issn1006-7736.2025.03.008

基于CycleGAN的无监督图像去雾网络

刘婷^*，董梦宇，郑凯，王琨，黄先阳，鹿香怡

作者信息 +

Unsupervised image dehazing network based on CycleGAN

LIU Ting^*, DONG Mengyu, ZHENG Kai, WANG Kun, HUANG Xianyang,LU Xiangyi

Author information +

文章历史 +

摘要

基于深度学习的去雾网络性能依赖大型丰富的成对数据集，然而真实环境中雾图与清晰图像的配对数据采集极为困难，现有研究多采用合成数据集训练，这导致网络在复杂真实场景中的泛化能力不足。为此，本文提出一种基于CycleGAN的无监督图像去雾网络，解决CycleGAN在去雾过程中易出现的颜色失真、伪影及去雾不彻底等问题。首先，设计残差密集连接模块构建编解码器架构，同时融入空间与通道残差注意力模块，并在跳跃连接部分引入注意力融合机制，实现特征提取与筛选过程的深度优化。其次，设计了新的损失函数，可有效平衡生成图像的视觉真实性与去雾准确性。实验结果表明，相比基线方法，该网络在室外合成数据集上去雾后的图片峰值信噪比值提升了22.71%，结构相似性值提升了6.49%，去雾后的图像质量得到提升；同时该网络还可以生成视觉效果逼真的雾图。

Abstract

The performance of deep learning-based dehazing networks is highly dependent on large and rich paired datasets. However, the collection of paired data consisting of hazy images and clear images in real-world environments is extremely difficult. As a result, existing studies mostly adopted synthetic datasets for training, which leads to insufficient generalization ability of the networks in complex real-world scenarios. Therefore, a CycleGAN-based unsupervised image dehazing network was proposed to solve the problems of color distortion, artifacts, and incomplete dehazing that were prone to occur during the dehazing process of CycleGAN. Firstly, a residual dense connection module was designed to construct the encoder-decoder architecture,while incorporating spatial and channel residual attention modules.An attention fusion mechanism was introduced in the skip connection section to achieve deep optimization of feature extraction and screening processes.Secondly, a new loss function was designed to effectively balance the visual authenticity and defogging accuracy of generated images.Experimental results show that, compared with baseline methods, the peak signal-to-noise ratio (PSNR) of the dehazed images of the designed network on outdoor synthetic datasets is improved by 22.71%, and the structural similarity (SSIM) value is increased by 6.49%. The image quality is improved after defogging, and the network can also generate visually realistic foggy images.

导出引用

刘婷, 董梦宇, 郑凯, 王琨, 黄先阳, 鹿香怡. 基于CycleGAN的无监督图像去雾网络[J]. 大连海事大学学报. 2025, 51(3): 74-84 https://doi.org/10.16411/j.cnki.issn1006-7736.2025.03.008

LIU Ting, DONG Mengyu, ZHENG Kai, WANG Kun, HUANG Xianyang, LU Xiangyi. Unsupervised image dehazing network based on CycleGAN[J]. Journal of Dalian Maritime University. 2025, 51(3): 74-84 https://doi.org/10.16411/j.cnki.issn1006-7736.2025.03.008

参考文献

[1]谢勇, 贾惠珍, 王同罕, 等. 图像去雾算法综述[J]. 计算机与数字工程, 2022, 50(12): 2765-2774.
XIE Y, JIA H Z, WANG T H, et al. A review of image dehazing algorithms[J]. Computer and Digital Engineering, 2022, 50(12): 2765-2774. (in Chinese)
[2]王科平, 杨艺, 费树岷. 雾霾图像清晰化算法综述[J]. 智能系统学报, 2023, 18(2): 217-230.
WANG K P, YANG Y, FEI S M. A review of haze image dehazing algorithms[J]. Journal of Intelligent Systems, 2023, 18(2): 217-230. (in Chinese)
[3]魏轶伦, 徐海文. 基于深度学习的图像去雾方法综述研究[J]. 信息技术与信息化, 2023(4): 214-216.
WEI Y L, XU H W. A review of image dehazing methods based on deep learning[J]. Information Technology and Informatization, 2023(4): 214-216. (in Chinese)
[4]CAI B L, XU X M, JIA K, et al. DehazeNet: an end-to-end system for single image haze removal[J]. IEEE Transactions on Image Processing, 2016, 25(11): 5187-5198.
[5]LI B Y, PENG X L, WANG Z Y, et al. AOD-Net: all-in-one dehazing network[C]//. IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 2017: 4780-4788.
[6]CHEN D D, HE M M, FAN Q N, et al. Gated context aggregation network for image dehazing and deraining[C]//. IEEE Winter Conference on Applications of Computer Vision (WACV), Waikoloa, HI, USA, 2019: 1375-1383.
[7]QIN X, WANG Z L, BAI Y C, et al. FFA-Net: feature fusion attention network for single image dehazing[C]//Proceedings of the AAAI Conference on Artificial Intelligence.New York:IEEE 2020:11908-11915.
[8]ZHU J Y, PARK T, ISOLA P, et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//IEEE International Conference on Computer Vision (ICCV).Venice:IEEE, 2017: 2242-2251.
[9]张成, 潘明阳, 高翊然, 等. 融合暗通道先验与循环生成对抗网络的航海图像去雾模型[J]. 大连海事大学学报, 2022, 48(4): 84-93.
ZHANG C, PAN M Y, GAO Y R, et al. A maritime image dehazing model based on the fusion of dark channel prior and CycleGAN[J]. Journal of Dalian Maritime University, 2022, 48(4): 84-93. (in Chinese)
[10]ENGIN D, GENÇ A, EKENEL H K. Cycle-dehaze: enhanced cyclegan for single image dehazing[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).Salt Lake City:IEEE, 2018: 938-9388.
[11]伏锦, 黄山. 基于改进循环生成对抗网络的图像去噪研究[J]. 计算机工程与应用, 2023, 59(17): 178-186.
FU J, HUANG S. Research on image denoising based on improved cycle generative adversarial network[J]. Computer Engineering and Applications, 2023, 59(17): 178-186. (in Chinese)
[12]姜鑫. 基于生成式对抗网络的图像去雾算法研究[D]. 北京: 中国科学院大学(中国科学院长春光学精密机械与物理研究所), 2022.
JIANG X. Research on image dehazing algorithm based on generative adversarial network[D]. Beijing: University of Chinese Academy of Sciences (Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences), 2022. (in Chinese)
[13]JIANG X, ZHAO C L, ZHU M, et al. Residual spatial and channel attention networks for single image dehazing[J]. Sensors, 2021, 21(23): 7922.
[14]周柏均. 复杂气象条件下船舶目标检测与识别算法研究[D]. 大连:大连海事大学, 2022.
ZHOU B J. Research on ship target detection and recognition algorithms under complex meteorological conditions[D]. Dalian: Dalian Maritime University, 2022. (in Chinese)
[15]ZHANG H, PATEL V M. Densely connected pyramid dehazing network[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City:IEEE,2018: 3194-3203.
[16]WANG C, CHEN R Q, LU Y, et al. Recurrent context aggregation network for single image dehazing[J]. IEEE Signal Processing Letters, 2021, 28: 419-423.
[17]HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Las Vegas:IEEE, 2016: 770-778.
[18]HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE,2018: 7132-7141.
[19]ZHANG H, SINDAGI V, PATEL V M. Multi-scale single image dehazing using perceptual pyramid deep network[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).Salt Lake City:IEEE,2018: 1015-1024.
[20]SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[C]//International Conference on Learning Representations Computational and Biological Learning Society.San Diego:IEEE,2015: 1-14.
[21]REZA A M. Realization of the contrast limited adaptive histogram equalization (CLAHE) for real-time image enhancement[J]. Journal of VLSI Signal Processing Systems for Signal, Image and Video Technology, 2004, 38(1): 35-44.
[22]HE K M, SUN J, TANG X O. Single image haze removal using dark channel prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 33(12): 2341-2353.
[23]FAN T, LI C, MA X, et al. An improved single image defogging method based on Retinex[C]//International Conference on Image, Vision and Computing (ICIVC). Chengdu:IEEE,2017: 410-413.
[24]HUYNH-THU Q, GHANBARI M. Scope of validity of PSNR in image/video quality assessment[J]. Electronics Letters, 2008, 44(13): 800-801.
[25]SILBERMAN N, HOIEM D, KOHLI P, et al. Indoor segmentation and support inference from rgbd images[C]//Proceedings of the 12th European Conference on Computer Vision+Volume Part V. Florence:IEEE,2012:746-760.
[26]ANCUTI C O, ANCUTI C, TIMOFTE R. NH-HAZE: An image dehazing benchmark with non-homogeneous hazy and haze-free images[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).Seattle:IEEE,2020: 444-445.
[27]ANCUTI C O, ANCUTI C, SBERT M, et al. Dense-haze: a benchmark for image dehazing with dense-haze and haze-free images[C]//IEEE International Conference on Image Processing (ICIP).Taipei:IEEE,2019: 1014-1018.
[28]ZHU J Y, PARK T, ISOLA P, et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//IEEE International Conference on Computer Vision (ICCV).Venice:IEEE,2017: 2242-2251.

基金

中国博士后科学基金资助项目（2019M661076）；大连市科技计划（重点）项目（2024JB11PT007）；国家自然科学基金资助项目（52071047）；国家重点研发计划（2021YFB3901501）；大连市杰出青年学者项目（2024RJ012）；中央高校基本科研业务费专项资金资助项目（3132023512）;水路交通控制全国重点实验室（大连）青年培育计划项目（3132025811）