融合特征增强与对比学习的自监督遥感图像去噪

何晓晴; 王志宝; 赵满; 陈良富; 毕秀丽

doi:10.11834/jrs.20254566

模型与方法 | 浏览量 : 0 下载量: 199 CSCD: 0

R-PDF
PDF
导出
分享
收藏
专辑

融合特征增强与对比学习的自监督遥感图像去噪
Self-supervised remote sensing image denoising algorithm based on feature enhancement and contrastive learning
2026年30卷第1期页码：198-212
收稿：2024-12-23，

纸质出版：2026-01-07
DOI： 10.11834/jrs.20254566
稿件说明：

移动端阅览

何晓晴，王志宝，赵满，陈良富，毕秀丽.2026.融合特征增强与对比学习的自监督遥感图像去噪.遥感学报，30（1）： 198-212 DOI： 10.11834/jrs.20254566.

He X Q，Wang Z B，Zhao M，Chen L F and Bi X L. 2026. Self-supervised remote sensing image denoising algorithm based on feature enhancement and contrastive learning. National Remote Sensing Bulletin， 30（1）：198-212 DOI： 10.11834/jrs.20254566.

摘要

由于遥感图像独特的获取方式，其采集过程容易受到噪声干扰，导致信息发生严重退化，而在现实世界中难以获取同一场景下的干净与噪声图像对。因此，自监督图像去噪成为了一个热门的研究方向。对于纹理复杂的遥感图像，现有的去噪方法存在细节丢失和背景模糊的问题。为此，本文提出了一种基于特征增强与对比学习的自监督遥感图像去噪算法，由去噪分支和对比分支两部分组成。在去噪分支中，首先设计了一个特征增强的卷积自编码器去噪网络，引入全局特征提取模块和注意力提取模块，分别获取不同尺度的浅层轮廓特征和局部细节特征；然后，利用动态特征增强模块扩展感受野以融合更多的空间结构信息；最后，在图像重建模块中通过动态自适应混合操作来保留深浅层的信息流。在对比分支中，该算法通过对比学习策略充分利用噪声图像携带的信息构建了新的对比感知损失，并联合重建损失和总变分损失来衡量去噪图像的平滑性和细节保持能力，有效减少了背景模糊的现象。最后，在NWPU-RESISC45与UC Merced Land Use这两个数据集上，将本文提出方法与其他去噪方法进行对比。结果表明本文方法在高斯噪声上的平均峰值信噪比提高了1.47—4.34 dB和2.06—4.95 dB，平均结构相似性提高了2.3%—11.8%和2.6%—11.5%。此外，在散斑噪声、条纹噪声以及真实噪声遥感图像上，利用本文方法也获得了很好的去噪效果。

Abstract

Optical remote sensing images are often disturbed by noise during transmission or storage

which affects the processing accuracy of downstream tasks. Thus

effectively removing noise from images is important. Obtaining clean and noisy image pairs in the same scene for optical remote sensing images is difficult

which is why many researchers have proposed self-supervised image denoising methods. Among them

convolutional autoencoders are a common self-supervised learning paradigm that requires no external labels. However

their current designs are often simplistic

limiting their ability to preserve fine textures in optical remote sensing images. Additionally

existing denoising methods often lack an effective perceptual loss function

which can result in over-smoothing after noise removal

with background blurring and artifacts.

To this end

this paper proposes a self-supervised remote sensing image denoising algorithm based on feature enhancement and contrastive learning

including two core parts: the denoising branch and the contrastive branch. First

in the denoising branch

this paper constructs a feature-enhanced convolutional autoencoder denoising network

which acquires shallow contour features at different scales by using a global feature extraction module. Then

a lightweight attention mechanism is introduced in the attention extraction module

which can effectively focus complex texture features from remote sensing images. Next

the dynamic enhancement module is used to dynamically expand the sensory field to incorporate more spatially structured information. Finally

a dynamic adaptive mix-up operation is introduced in the image reconstruction module to encourage the shallow feature information in the downsampled part to flow adaptively to the deeper features in the upsampled part and thus effectively preserve the detailed features. In addition

in the contrastive branch

this paper utilizes the feature information carried by the noisy images to construct positive and negative sample pairs by using different data enhancement strategies to compute a new contrastive perceptual loss. This loss forms two opposite forces in the feature space: pulling the denoised image closer to the clean image and pushing it away from the noisy one. Meanwhile

total variation loss is applied to reduce pixel variations and thus better preserve edge details. Finally

reconstruction loss

total variation loss

and contrastive perception loss are used as a joint loss function to guide the network training to achieve the best denoising effect.

Experimental results show that on the NWPU-RESISC45 and UC Merced Land Use datasets

the proposed method improves the average PSNR on Gaussian noise by 1.47—4.34 and 2.06—4.95 dB

and the average SSIM by 2.3%—11.8% and 2.6%—11.5% compared with other denoising methods. In addition

the proposed method achieves satisfactory denoising results on Speckle noise

Stripe noise

and real noisy remote sensing images.

Whether on synthetic or real noise experiments

the proposed method can retain richer detail features after removing noise and avoid background blurring and artefacts. In addition

the proposed method has good generalizability and can handle Gaussian noise

speckle noise

and stripe noise.

关键词

Keywords

references

Chen T , Kornblith S , Norouzi M and Hinton G . 2020 . A simple framework for contrastive learning of visual representations //Proceedings of the 37th International Conference on Machine Learning. [s.l. ] : JMLR .org: 149 [DOI: 10.48550/arXiv.2002.05709 ]

Cheng G , Han J W and Lu X Q . 2017 . Remote sensing image scene classification: benchmark and state of the art . Proceedings of the IEEE , 105 ( 10 ): 1865 - 1883 [ DOI: 10.1109/JPROC.2017.2675998 http://dx.doi.org/10.1109/JPROC.2017.2675998 ]

Cheng L B , Du W L , Li Z and Jia X N . 2025 . AFEV-INet: Adaptive feature extraction variational interactive network for remote sensing image denoising . Multimedia Systems , 31 ( 2 ): 90 [ DOI: 10.1007/s00530-025-01680-1 http://dx.doi.org/10.1007/s00530-025-01680-1 ]

Dabov K , Foi A , Katkovnik V and Egiazarian K . 2007 . Image denoising by sparse 3-D transform-domain collaborative filtering . IEEE Transactions on Image Processing , 16 ( 8 ): 2080 - 2095 [ DOI: 10.1109/TIP.2007.901238 http://dx.doi.org/10.1109/TIP.2007.901238 ]

Feng Y C , Jiang J W , Xu H H and Zheng J W . 2023 . Change detection on remote sensing images using dual-branch multilevel intertemporal network . IEEE Transactions on Geoscience and Remote Sensing , 61 : 4401015 [ DOI: 10.1109/TGRS.2023.3241257 http://dx.doi.org/10.1109/TGRS.2023.3241257 ]

Geng H , Gu Z X , Weng L Y and Yu J . 2023 . A non-local image denoising method based on TV-L1 with variable exponents // Proceedings Volume 12705, Fourteenth International Conference on Graphics and Image Processing (ICGIP 2022) . Nanjing : SPIE: 707 - 715 [ DOI: 10.1117/12.2680050 http://dx.doi.org/10.1117/12.2680050 ]

Han L T , Zhao Y C , Lv H Y , Zhang Y S , Liu H L and Bi G L . 2022 . Remote sensing image denoising based on deep and shallow feature fusion and attention mechanism . Remote Sensing , 14 ( 5 ): 1243 [ DOI: 10.3390/rs14051243 http://dx.doi.org/10.3390/rs14051243 ]

Huang T , Li S J , Jia X , Lu H C and Liu J Z . 2022 . Neighbor2Neighbor: a self-supervised framework for deep image denoising . IEEE Transactions on Image Processing , 31 : 4023 - 4038 [ DOI: 10.1109/TIP.2022.3176533 http://dx.doi.org/10.1109/TIP.2022.3176533 ]

Jia H B , Yin Q B and Lu M Y . 2022 . Blind-noise image denoising with block-matching domain transformation filtering and improved guided filtering . Scientific Reports , 12 ( 1 ): 16195 [ DOI: 10.1038/s41598-022-20578-w http://dx.doi.org/10.1038/s41598-022-20578-w ]

Lehtinen J , Munkberg J , Hasselgren J , Laine S , Karras T , Ai-ttala M and Aila T . 2018 . Noise2Noise: learning image restoration without clean data // Proceedings of the 35th International Conference on Machine Learning . Stockholm : PMLR: 2965 - 2974 [ DOI: 10.48550/arXiv.1803.04189 http://dx.doi.org/10.48550/arXiv.1803.04189 ]

Li J , Lin L P , He M G , He J , Yuan Q Q and Shen H F . 2024 . Sentinel-1 dual-polarization SAR images despeckling network based on unsupervised learning . IEEE Transactions on Geoscience and Remote Sensing , 62 : 1 - 15 [ DOI: 10.1109/TGRS.2024.3404405 http://dx.doi.org/10.1109/TGRS.2024.3404405 ]

Li J Y , Zhang Z L , Liu X Y , Feng C Y , Wang X T and Lei L . 2023 . Spatially adaptive self-supervised learning for real-world image denoising // Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Vancouver : IEEE: 9914 - 9924 [ DOI: 10.1109/CVPR52729.2023.00956 http://dx.doi.org/10.1109/CVPR52729.2023.00956 ]

Liu M H , Jiang W Z , Liu W F , Tao D P and Liu B D . 2023a . Dynamic adaptive attention-guided self-supervised single remote-sensing image denoising . IEEE Transactions on Geoscience and Remote Sensing , 61 : 4704511 [ DOI: 10.1109/TGRS.2023.3299636 http://dx.doi.org/10.1109/TGRS.2023.3299636 ]

Liu Y K , Wan B W , Shi D M and Cheng X C . 2023b . Generative recorrupted-to-recorrupted: an unsupervised image denoising network for arbitrary noise distribution . Remote Sensing , 15 ( 2 ): 364 [ DOI: 10.3390/rs15020364 http://dx.doi.org/10.3390/rs15020364 ]

Luo Z Z , Wang X Y , Pellikka P , Heiskanen J and Zhong Y F . 2024 . Unsupervised adaptation learning for real multiplatform hyperspectral image denoising . IEEE Transactions on Cybernetics , 54 ( 10 ): 5781 - 5794 [ DOI: 10.1109/TCYB.2024.3412270 http://dx.doi.org/10.1109/TCYB.2024.3412270 ]

Mittal A , Moorthy A K and Bovik A C . 2012 . No-reference image quality assessment in the spatial domain . IEEE Transactions on Image Processing , 21 ( 12 ): 4695 - 4708 [ DOI: 10.1109/TIP.2012.2214050 http://dx.doi.org/10.1109/TIP.2012.2214050 ]

Mittal A , Soundararajan R and Bovik A C . 2013 . Making a “completely blind” image quality analyzer . IEEE Signal Processing Letters , 20 ( 3 ): 209 - 212 [ DOI: 10.1109/LSP.2012.2227726 http://dx.doi.org/10.1109/LSP.2012.2227726 ]

Moran N , Schmidt D , Zhong Y and Coady P . 2020 . Noisier2Noise: learning to denoise from unpaired noisy data // Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Seattle : IEEE: 12061 - 12069 [ DOI: 10.1109/CVPR42600.2020.01208 http://dx.doi.org/10.1109/CVPR42600.2020.01208 ]

Pan Y Z , Liu X , Liao X Y , Cao Y Z H and Ren C . 2023 . Random sub-samples generation for self-supervised real image denoising // Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision . Paris : IEEE: 12116 - 12125 [ DOI: 10.1109/ICCV51070.2023.01116 http://dx.doi.org/10.1109/ICCV51070.2023.01116 ]

Tao C , Qi J , Guo M N , Zhu Q and Li H F . 2023 . Self-supervised remote sensing feature learning: learning paradigms, challenges, and future works . IEEE Transactions on Geoscience and Remote Sensing , 61 : 1 - 26 [ DOI: 10.1109/TGRS.2023.3276853 http://dx.doi.org/10.1109/TGRS.2023.3276853 ]

Thai D H , Fei X Q , Le M T , Züfle A and Wessels K . 2023 . Riesz-Quincunx-UNet variational autoencoder for unsupervised satellite image denoising . IEEE Transactions on Geoscience and Remote Sensing , 61 : 5404519 [ DOI: 10.1109/TGRS.2023.3291309 http://dx.doi.org/10.1109/TGRS.2023.3291309 ]

Wang P Q , Chen P F , Yuan Y , Liu D , Huang Z H , Hou X D and Cottrell G . 2018 . Understanding convolution for semantic segmentation // 2018 IEEE Winter Conference on Applications of Computer Vision (WACV) . Lake Tahoe : IEEE: 1451 - 1460 [ DOI: 10.1109/WACV.2018.00163 http://dx.doi.org/10.1109/WACV.2018.00163 ]

Wang Z , Wei J Z , Wang Y and Li Q . 2024c . Remote sensing image denoising algorithm based on improved transformer network // Proceedings Volume 13182, 2024 International Conference on Optoelectronic Information and Optical Engineering (OIOE 2024) . Kunming : SPIE: 169 - 176 [ DOI: 10.1117/12.3030732 http://dx.doi.org/10.1117/12.3030732 ]

Wang Z B , Chang H , Bai L , Chen L F and Bi X L . 2024a . A creative weak supervised semantic segmentation for remote sensing images . IEEE Transactions on Geoscience and Remote Sensing , 62 : 1 - 13 [ DOI: 10.1109/TGRS.2024.3477749 http://dx.doi.org/10.1109/TGRS.2024.3477749 ]

Wang Z B , He X Q , Xiao B , Chen L F and Bi X L . 2024b . RSID-CR: remote sensing image denoising based on contrastive learning . IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 17 : 18784 - 18799 [ DOI: 10.1109/JSTARS.2024.3476566 http://dx.doi.org/10.1109/JSTARS.2024.3476566 ]

Woo S , Park J , Lee J Y and Kweon I S . 2018 . CBAM: convolutional block attention module // Proceedings of the 15th European Conference on Computer Vision – ECCV 2018 . Munich : Springer: 3 - 19 [ DOI: 10.1007/978-3-030-01234-2_1 http://dx.doi.org/10.1007/978-3-030-01234-2_1 ]

Xia G S , Hu J W , Hu F , Shi B G , Bai X , Zhong Y F , Zhang L P and Lu X Q . ( 2017 ). AID: a benchmark data set for performance evaluation of aerial scene classification . IEEE Transactions on Geoscience and Remote Sensing , 55 ( 7 ): 3965 - 3981 [ DOI: 10.1109/TGRS.2017.2685945 http://dx.doi.org/10.1109/TGRS.2017.2685945 ]

Yang L X , Zhang R Y , Li L D and Xie X H . 2021 . SimAM: a simple, parameter-free attention module for convolutional neural networks //Proceedings of the 38th International Conference on Machine Learning. [s.l. ] : PMLR: 11863 - 11874

Yang Y and Newsam S . 2010 . Bag-of-visual-words and spatial extensions for land-use classification // Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems . San Jose : Association for Computing Machinery: 270 - 279 [ DOI: 10.1145/1869790.1869829 http://dx.doi.org/10.1145/1869790.1869829 ]

Yu B W , Zhou Y F , Liu X and Wang X G . 2022 . A non-local means based multiplicative denoising method for image processing // Proceedings of the 6th International Technical Conference on Advances in Computing, Control and Industrial Engineering (CCIE 2021) . Singapore : Springer: 763 - 773 [ DOI: 10.1007/978-981-19-3927-3_75 http://dx.doi.org/10.1007/978-981-19-3927-3_75 ]

Zhang D , Zhou F F , Jiang Y W and Fu Z M . 2023a . MM-BSN: self-supervised image denoising for real-world with multi-mask based on blind-spot network // Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) . Vancouver : IEEE: 4189 - 4198 [ DOI: 10.1109/CVPRW59228.2023.00441 http://dx.doi.org/10.1109/CVPRW59228.2023.00441 ]

Zhang J , Lu M X , Li J K , Xu D Y , Huang W X and Shi X P . 2024 . Residual dense convolutional autoencoder for high noise image denoising . Computer Science , 51 ( 6 A): 230400073

张杰 , 卢淼鑫 , 李嘉康 , 徐大勇 , 黄雯潇 , 史小平 . 2024 . 基于残差密集卷积自编码的高噪声图像去噪方法 . 计算机科学 , 51 ( 6 A): 230400073 [ DOI: 10.11896/jsjkx.230400073 http://dx.doi.org/10.11896/jsjkx.230400073 ]

Zhang K , Zuo W M , Chen Y J , Meng D Y and Zhang L . 2017 . Beyond a Gaussian denoiser: residual learning of deep CNN for image denoising . IEEE Transactions on Image Processing , 26 ( 7 ): 3142 - 3155 [ DOI: 10.1109/TIP.2017.2662206 http://dx.doi.org/10.1109/TIP.2017.2662206 ]

Zhang K , Zuo W M and Zhang L . 2018 . FFDNet: toward a fast and flexible solution for CNN-based image denoising . IEEE Transactions on Image Processing , 27 ( 9 ): 4608 - 4622 [ DOI: 10.1109/TIP.2018.2839891 http://dx.doi.org/10.1109/TIP.2018.2839891 ]

Zhang Y , Bai L , Wang Z B , Fan M , Jurek-Loughrey A , Zhang Y Q , Zhang Y , Zhao M and Chen L F . 2023b . Oil well detection under occlusion in remote sensing images using the improved YOLOv5 model . Remote Sensing , 15 ( 24 ): 5788 [ DOI: 10.3390/rs15245788 http://dx.doi.org/10.3390/rs15245788 ]