Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired Training
Speckle suppression is a crucial preliminary step for synthetic aperture radar (SAR) image processing. Supervised despeckling approaches trained on synthetic datasets usually perform poorly in practice due to the unavailability of clean SAR images. Besides, the spatial correlation of speckle is rare...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2023-01-01
|
| Series: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10294190/ |
| _version_ | 1827362032565878784 |
|---|---|
| author | Can Wang Rongyao Zheng Jingzhen Zhu Xingkun He Xiwen Li |
| author_facet | Can Wang Rongyao Zheng Jingzhen Zhu Xingkun He Xiwen Li |
| author_sort | Can Wang |
| collection | DOAJ |
| description | Speckle suppression is a crucial preliminary step for synthetic aperture radar (SAR) image processing. Supervised despeckling approaches trained on synthetic datasets usually perform poorly in practice due to the unavailability of clean SAR images. Besides, the spatial correlation of speckle is rarely considered in many methods based on the fully developed speckle assumption. In this article, we propose an unsupervised despeckling method to address these issues by combining online speckle generation and unpaired training. The method consists of two branches: the stop-gradient branch and the unpaired branch. First, the stop-gradient branch learns to generate the spatially correlated speckle. Then, the unpaired branch combines the generated speckle with the unpaired optical image to form pairs of training data for network parameter updates. More specifically, in order to generate the more realistic speckle in the stop-gradient branch, we design a speckle correction module with three SAR speckle priors: the threshold prior, the unit mean prior, and the correlation prior coupled with the weighted patch-shuffle. In the unpaired training, a hybrid loss function is employed, which takes spatial smoothness and detail protection into consideration. Afterward, we combine the stop-gradient branch with the unpaired branch by the Siamese network to achieve alternate optimization of speckle generation and speckle removal. Finally, the optimization process in our method is analyzed theoretically. Qualitative and quantitative experiments demonstrate that the proposed method is comparable to the supervised despeckling approaches on synthetic datasets and outperforms several state-of-the-art unsupervised methods on real SAR datasets. |
| first_indexed | 2024-03-08T07:19:07Z |
| format | Article |
| id | doaj.art-2101e9e07fa34cfaa366338ea8706490 |
| institution | Directory Open Access Journal |
| issn | 2151-1535 |
| language | English |
| last_indexed | 2024-03-08T07:19:07Z |
| publishDate | 2023-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| spelling | doaj.art-2101e9e07fa34cfaa366338ea87064902024-02-03T00:01:41ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing2151-15352023-01-0116101751019010.1109/JSTARS.2023.332718010294190Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired TrainingCan Wang0https://orcid.org/0000-0002-5670-1493Rongyao Zheng1https://orcid.org/0000-0003-0831-0719Jingzhen Zhu2https://orcid.org/0000-0002-0204-4889Xingkun He3https://orcid.org/0000-0001-6701-3808Xiwen Li4https://orcid.org/0000-0002-9736-2037State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, ChinaState Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, ChinaState Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, ChinaState Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, ChinaState Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, ChinaSpeckle suppression is a crucial preliminary step for synthetic aperture radar (SAR) image processing. Supervised despeckling approaches trained on synthetic datasets usually perform poorly in practice due to the unavailability of clean SAR images. Besides, the spatial correlation of speckle is rarely considered in many methods based on the fully developed speckle assumption. In this article, we propose an unsupervised despeckling method to address these issues by combining online speckle generation and unpaired training. The method consists of two branches: the stop-gradient branch and the unpaired branch. First, the stop-gradient branch learns to generate the spatially correlated speckle. Then, the unpaired branch combines the generated speckle with the unpaired optical image to form pairs of training data for network parameter updates. More specifically, in order to generate the more realistic speckle in the stop-gradient branch, we design a speckle correction module with three SAR speckle priors: the threshold prior, the unit mean prior, and the correlation prior coupled with the weighted patch-shuffle. In the unpaired training, a hybrid loss function is employed, which takes spatial smoothness and detail protection into consideration. Afterward, we combine the stop-gradient branch with the unpaired branch by the Siamese network to achieve alternate optimization of speckle generation and speckle removal. Finally, the optimization process in our method is analyzed theoretically. Qualitative and quantitative experiments demonstrate that the proposed method is comparable to the supervised despeckling approaches on synthetic datasets and outperforms several state-of-the-art unsupervised methods on real SAR datasets.https://ieeexplore.ieee.org/document/10294190/Deep learningimage despecklingSiamese networksynthetic aperture radar (SAR)unsupervised training |
| spellingShingle | Can Wang Rongyao Zheng Jingzhen Zhu Xingkun He Xiwen Li Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired Training IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Deep learning image despeckling Siamese network synthetic aperture radar (SAR) unsupervised training |
| title | Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired Training |
| title_full | Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired Training |
| title_fullStr | Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired Training |
| title_full_unstemmed | Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired Training |
| title_short | Unsupervised SAR Despeckling by Combining Online Speckle Generation and Unpaired Training |
| title_sort | unsupervised sar despeckling by combining online speckle generation and unpaired training |
| topic | Deep learning image despeckling Siamese network synthetic aperture radar (SAR) unsupervised training |
| url | https://ieeexplore.ieee.org/document/10294190/ |
| work_keys_str_mv | AT canwang unsupervisedsardespecklingbycombiningonlinespecklegenerationandunpairedtraining AT rongyaozheng unsupervisedsardespecklingbycombiningonlinespecklegenerationandunpairedtraining AT jingzhenzhu unsupervisedsardespecklingbycombiningonlinespecklegenerationandunpairedtraining AT xingkunhe unsupervisedsardespecklingbycombiningonlinespecklegenerationandunpairedtraining AT xiwenli unsupervisedsardespecklingbycombiningonlinespecklegenerationandunpairedtraining |