青岛大学学报（自然科学版）2024，Vol.37Issue(2)：89-97,125,10.DOI:10.3969/j.issn.1006-1037.2024.02.15
基于循环生成对抗网络和Transformer的单幅图像去雾算法
Single Image Dehazing Algorithm Based on Cycle Generative Adversarial Networks and Transformer
摘要
Abstract
Aiming at the problem of overfitting in traditional dehazing algorithms trained on paired data-sets,a non-paired image dehazing network model based on density and depth decomposition was improved with a self-enhancing scaling network.Introducing the Transformer mechanism and deeply integrating it with deep convolutional neural networks for network module deep fusion,a CT-Nets image dehazing algo-rithm based on cycle generative adversarial networks and Transformers trained on unpaired datasets was proposed.The depth information and scattering coefficient eigenvalues of the input image were extracted,and the atmospheric scattering model was used to restore the real fog concentration information in different scenes as much as possible to improve the subjective visual quality of the defogged image.Based on Swin-Transformer,a self-enhancing refinement layer to obtain finer-grained information was designed to im-prove the generalization ability of the model and the authenticity of the final predicted image.The experi-mental results show that compared to the dehazing via decomposing transmission map into density and depth network model,the peak signal-to-noise ratio and structural similarity of the CT-Nets image dehaz-ing algorithm are improved by 4%and 4.1%,respectively.关键词
深度学习/单幅图像去雾/自监督网络/循环生成对抗网络Key words
deep learning/single image dehazing/self-supervised network/cycle generative adversarial分类
信息技术与安全科学引用本文复制引用
王博,魏伟波,张为栋,潘振宽,李明,李金函..基于循环生成对抗网络和Transformer的单幅图像去雾算法[J].青岛大学学报（自然科学版）,2024,37(2):89-97,125,10.基金项目
山东省自然科学基金(批准号:ZR2020 QF033)资助. （批准号:ZR2020 QF033）
