{"title":"基于条件GAN的多传感器遥感图像超分辨率","authors":"Junwei Wang, Kun Gao, Zhenzhou Zhang, Chong Ni, Zibo Hu, Dayu Chen, Qiong Wu","doi":"10.34133/2021/9829706","DOIUrl":null,"url":null,"abstract":"Despite the promising performance on benchmark datasets that deep convolutional neural networks have exhibited in single image super-resolution (SISR), there are two underlying limitations to existing methods. First, current supervised learning-based SISR methods for remote sensing satellite imagery do not use paired real sensor data, instead operating on simulated high-resolution (HR) and low-resolution (LR) image-pairs (typically HR images with their bicubic-degraded LR counterparts), which often yield poor performance on real-world LR images. Second, SISR is an ill-posed problem, and the super-resolved image from discriminatively trained networks with lp norm loss is an average of the infinite possible HR images, thus, always has low perceptual quality. Though this issue can be mitigated by generative adversarial network (GAN), it is still hard to search in the whole solution-space and find the best solution. In this paper, we focus on real-world application and introduce a new multisensor dataset for real-world remote sensing satellite imagery super-resolution. In addition, we propose a novel conditional GAN scheme for SISR task which can further reduce the solution-space. Therefore, the super-resolved images have not only high fidelity, but high perceptual quality as well. Extensive experiments demonstrate that networks trained on the introduced dataset can obtain better performances than those trained on simulated data. Additionally, the proposed conditional GAN scheme can achieve better perceptual quality while obtaining comparable fidelity over the state-of-the-art methods.","PeriodicalId":38304,"journal":{"name":"遥感学报","volume":" ","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2021-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"8","resultStr":"{\"title\":\"Multisensor Remote Sensing Imagery Super-Resolution with Conditional GAN\",\"authors\":\"Junwei Wang, Kun Gao, Zhenzhou Zhang, Chong Ni, Zibo Hu, Dayu Chen, Qiong Wu\",\"doi\":\"10.34133/2021/9829706\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Despite the promising performance on benchmark datasets that deep convolutional neural networks have exhibited in single image super-resolution (SISR), there are two underlying limitations to existing methods. First, current supervised learning-based SISR methods for remote sensing satellite imagery do not use paired real sensor data, instead operating on simulated high-resolution (HR) and low-resolution (LR) image-pairs (typically HR images with their bicubic-degraded LR counterparts), which often yield poor performance on real-world LR images. Second, SISR is an ill-posed problem, and the super-resolved image from discriminatively trained networks with lp norm loss is an average of the infinite possible HR images, thus, always has low perceptual quality. Though this issue can be mitigated by generative adversarial network (GAN), it is still hard to search in the whole solution-space and find the best solution. In this paper, we focus on real-world application and introduce a new multisensor dataset for real-world remote sensing satellite imagery super-resolution. In addition, we propose a novel conditional GAN scheme for SISR task which can further reduce the solution-space. Therefore, the super-resolved images have not only high fidelity, but high perceptual quality as well. Extensive experiments demonstrate that networks trained on the introduced dataset can obtain better performances than those trained on simulated data. Additionally, the proposed conditional GAN scheme can achieve better perceptual quality while obtaining comparable fidelity over the state-of-the-art methods.\",\"PeriodicalId\":38304,\"journal\":{\"name\":\"遥感学报\",\"volume\":\" \",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2021-09-08\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"8\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"遥感学报\",\"FirstCategoryId\":\"1089\",\"ListUrlMain\":\"https://doi.org/10.34133/2021/9829706\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"遥感学报","FirstCategoryId":"1089","ListUrlMain":"https://doi.org/10.34133/2021/9829706","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Multisensor Remote Sensing Imagery Super-Resolution with Conditional GAN
Despite the promising performance on benchmark datasets that deep convolutional neural networks have exhibited in single image super-resolution (SISR), there are two underlying limitations to existing methods. First, current supervised learning-based SISR methods for remote sensing satellite imagery do not use paired real sensor data, instead operating on simulated high-resolution (HR) and low-resolution (LR) image-pairs (typically HR images with their bicubic-degraded LR counterparts), which often yield poor performance on real-world LR images. Second, SISR is an ill-posed problem, and the super-resolved image from discriminatively trained networks with lp norm loss is an average of the infinite possible HR images, thus, always has low perceptual quality. Though this issue can be mitigated by generative adversarial network (GAN), it is still hard to search in the whole solution-space and find the best solution. In this paper, we focus on real-world application and introduce a new multisensor dataset for real-world remote sensing satellite imagery super-resolution. In addition, we propose a novel conditional GAN scheme for SISR task which can further reduce the solution-space. Therefore, the super-resolved images have not only high fidelity, but high perceptual quality as well. Extensive experiments demonstrate that networks trained on the introduced dataset can obtain better performances than those trained on simulated data. Additionally, the proposed conditional GAN scheme can achieve better perceptual quality while obtaining comparable fidelity over the state-of-the-art methods.
遥感学报Social Sciences-Geography, Planning and Development
CiteScore
3.60
自引率
0.00%
发文量
3200
期刊介绍:
The predecessor of Journal of Remote Sensing is Remote Sensing of Environment, which was founded in 1986. It was born in the beginning of China's remote sensing career and is the first remote sensing journal that has grown up with the development of China's remote sensing career. Since its inception, the Journal of Remote Sensing has published a large number of the latest scientific research results in China and the results of nationally-supported research projects in the light of the priorities and needs of China's remote sensing endeavours at different times, playing a great role in the development of remote sensing science and technology and the cultivation of talents in China, and becoming the most influential academic journal in the field of remote sensing and geographic information science in China.
As the only national comprehensive academic journal in the field of remote sensing in China, Journal of Remote Sensing is dedicated to reporting the research reports, stage-by-stage research briefs and high-level reviews in the field of remote sensing and its related disciplines with international and domestic advanced level. It focuses on new concepts, results and progress in this field. It covers the basic theories of remote sensing, the development of remote sensing technology and the application of remote sensing in the fields of agriculture, forestry, hydrology, geology, mining, oceanography, mapping and other resource and environmental fields as well as in disaster monitoring, research on geographic information systems (GIS), and the integration of remote sensing with GIS and the Global Navigation Satellite System (GNSS) and its applications.