Bridging street view coverage disparities through geographic identity preserving generation from satellite view
文献类型：期刊论文
| 作者 | Li, Zongrong4,5; Zhang, Fan3; Dai, Shaoqing1,2; Zhao, Wufan5 |
| 刊名 | ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING
 |
| 出版日期 | 2026-06-01 |
| 卷号 | 236页码:622-639 |
| 关键词 | Cross-view image generation Satellite-to-street view synthesis Geographic identity Urban data equity |
| ISSN号 | 0924-2716 |
| DOI | 10.1016/j.isprsjprs.2026.03.049 |
| 产权排序 | 5 |
| 文献子类 | Article |
| 英文摘要 | Street view imagery (SVI) provides a human-centric view of urban environments and is widely used for analyzing greenery, mobility, socioeconomics conditions, health outcomes, and safety perception. However, its coverage is highly uneven, with developing regions systematically underrepresented, limiting the inclusiveness of SVI-based analytics. To address this, we propose GeoIdentity-Sat2Street, a geographic identity preserving framework, that leverages satellite imagery to expand SVI coverage. Our framework first applies a polar-transformation conditional GAN to synthesize plausible street view perspectives, then refines them with a diffusion-based generator conditioned on semantic captions, location metadata, and structural priors. This design enforces geometric consistency while explicitly preserving geographic identity, an overlooked but critical aspect of urban distinctiveness. We conduct a comprehensive evaluation against baselines, covering generation quality, geographic identity preservation, and global scalability. First, on classic cross-view benchmarks (CVUSA and CVACT), our method achieves the highest SSIM (0.427/0.537), lowest LPIPS (0.345/0.317), and best mIoU (0.054). Next, to assess geographic identity fidelity, we introduce MultiCities Dataset, a benchmark of 50,000 paired satellite-street-view images across five cities on five continents. Our method achieves the highest Silhouette Score (0.222), lowest inter-city variance (0.031), and clearly separated clusters in t-SNE, demonstrating superior preservation of regional visual identity; GPT-based evaluation further confirms realism and semantic alignment (score 10.587-10.824). Finally, we apply our model to Kathmandu, Nepal achieving about a 28% improvement in usable street-view coverage, from 72% to dense roadside coverage. Overall, this work highlights the proposed identity preserving two-step pipeline as central to equitable SVI generation and provides a scalable framework toward globally representative urban analytics. |
| URL标识 | 查看原文 |
| WOS研究方向 | Physical Geography ; Geology ; Remote Sensing ; Imaging Science & Photographic Technology |
| 语种 | 英语 |
| WOS记录号 | WOS:001745742800001 |
| 出版者 | ELSEVIER |
| 源URL | [http://ir.igsnrr.ac.cn/handle/311030/221545]  |
| 专题 | 资源与环境信息系统国家重点实验室_外文论文 |
| 通讯作者 | Dai, Shaoqing; Zhao, Wufan |
| 作者单位 | 1.Chinese Acad Sci, Inst Geog Sci & Nat Resources Res, State Key Lab Resources & Environm Informat Syst, Beijing, Peoples R China 2.Wuhan Univ, Sch Resource & Environm Sci, Wuhan, Peoples R China; 3.Peking Univ, Inst Remote Sensing & Geog Informat Syst, Sch Earth & Space Sci, Beijing, Peoples R China; 4.Univ Southern Calif, Spatial Sci Inst, Los Angeles, CA USA; 5.Hong Kong Univ Sci & Technol Guangzhou, Thrust Urban Governance & Design, Guangzhou, Peoples R China; |
| 推荐引用方式 GB/T 7714 | Li, Zongrong,Zhang, Fan,Dai, Shaoqing,et al. Bridging street view coverage disparities through geographic identity preserving generation from satellite view[J]. ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING,2026,236:622-639. |
| APA | Li, Zongrong,Zhang, Fan,Dai, Shaoqing,&Zhao, Wufan.(2026).Bridging street view coverage disparities through geographic identity preserving generation from satellite view.ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING,236,622-639. |
| MLA | Li, Zongrong,et al."Bridging street view coverage disparities through geographic identity preserving generation from satellite view".ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING 236(2026):622-639. |
入库方式： OAI收割
来源：地理科学与资源研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。