论文与情报流

聚合权威论文、预印本与机构内容源,支持按主题、来源与关键词筛选,面向实验室持续开展科研跟踪与资料沉淀。

筛选器
当前来源
当前展示 24 条,共 6130
论文列表
首页/论文
论文
International Journal of Geographical Information Science
PublisherJournal
Multimodal
基于遥感与社交媒体数据的高分辨率人口构成制图:一种多模态、多输出建模方法
High-resolution mapping of demographic compositions based on remote sensing and social media data: a multimodal, multi-output modeling approach

精细尺度的人口构成(如性别与年龄结构)认知,对公共卫生、城市规划及市场营销等多个领域具有基础性意义。然而,现有精细尺度人口产品大多未能刻画行政单元内部人口构成的空间异质性。本研究首次提出一种新方法,利用人口普查数据、遥感影像及带地理标签的社交媒体数据等多模态数据,在100 × 100米网格单元上估算完整的人口性别与年龄结构。该方法从遥感影像中提取关键环境变量(如人工灯光强度与植被指数),并从社交媒体文本中提取与人口特征相关的变量,通过考虑多个人口属性间的相关性,采用多输出随机森林模型,估计上述变量与普查所得人口构成(包括性别比与各年龄组占比)之间的关联。最终产品的性别比(每100名女性对应的男性数量)均方根误差(RMSE)为6.18;0–14岁、15–59岁及≥60岁年龄组占总人口比例的RMSE分别为1.89%、3.82%和3.40%。该方法具备开展精细化人口建模的潜力,可拓展至其他社会人口学因子,并惠及多个学科领域。

Ge Qiu Yuchen Li Shaoqing Dai Kun Qin Zhanpeng Wang Yaolin Liu Jiping Liu Peng Jia a School of Resource and Environmental Sciences, Center for Metabolic and Panvascular Diseases, Wuhan University, Wuhan, Chinab Medical Remote Sensing Information Research Institute (MRSIRI), Renmin Hospital of Wuhan University, Wuhan, Chinac Chinese Academy of Surveying and Mapping, Beijing, Chinad International Institute of Spatial Lifecourse Health (ISLE), Wuhan University, Wuhan, Chinae School of Geography, University of Leeds, Leeds, UKf Duke Kunshan University, Kunshan, ChinaGe Qiu received his Ph.D. in Cartography and Geographic Information Systems from the School of Resource and Environmental Sciences, Wuhan University. His research interests are multi-source geospatial data fusion, spatial modeling and mapping, GeoAI, and health geography. His primary contributions to this study included conceptualization, methodology, investigation, formal analysis, software, validation, visualization, data curation, writing – original draft, and writing – review and editing.Yuchen Li is a Lecturer at the School of Geography, University of Leeds. His research interests are spatial analysis and modeling, spatial data mining, spatial statistics, transportation simulation, and public health modeling. His primary contributions to this study included methodology, validation, writing – original draft, and writing – review and editing.Shaoqing Dai is a postdoctoral researcher at the School of Resource and Environmental Sciences, Wuhan University. His research interests are health geography, spatial statistics, GeoAI, and urban visual intelligence. His primary contributions to this study included investigation, validation, visualization, and writing – review and editing.Kun Qin is a Ph.D. candidate at the School of Resource and Environmental Sciences, Wuhan University. His research interests include spatial modeling and mapping, spatiotemporal analysis, and health geography. His primary contributions to this study included software, visualization, data curation, and writing – review and editing.Zhanpeng Wang received his Ph.D. in Cartography and Geographic Information Systems from the School of Resource and Environmental Sciences, Wuhan University. His research interests are remote sensing, GeoAI, and health geography. His primary contributions to this study included investigation, visualization, data curation, and writing – review and editing.Yaolin Liu is a Professor at the School of Resource and Environmental Sciences, Wuhan University, and currently serves as the Chancellor of Duke Kunshan University. His research focuses on the application of geospatial technologies in land resource surveying, evaluation, planning, and management. His primary contributions to this study included conceptualization, validation, resources, supervision, project administration, and writing – review and editing.Jiping Liu is a Research Fellow and Deputy Director General at the Chinese Academy of Surveying and Mapping. His research interests are spatiotemporal big data analysis and mining, governmental geospatial information services, emergency geospatial information services, and 3D real-world modeling. His primary contributions to this study included conceptualization, investigation, validation, resources, supervision, funding acquisition, project administration, and writing – review and editing.Peng Jia is the founding director of the International Institute of Spatial Lifecourse Health (ISLE) and the Deputy Director of Medical Remote Sensing Information Research Institute (MRSIRI) at Renmin Hospital of Wuhan University. His research interests include health geography, spatial epidemiology, environmental health, and spatial science and technology. His primary contributions to this study included conceptualization, validation, resources, supervision, funding acquisition, project administration, and writing – review and editing.
2026/07/27
论文
International Journal of Geographical Information Science
PublisherJournal
GeoAI
GEEToker:一种与模型无关的分词器,可提升大语言模型在 Google Earth Engine 中基于 JavaScript 的地理空间代码生成能力
GEEToker: you can boost LLMs’ JavaScript-based geospatial code generation for Google Earth Engine with a model-agnostic tokenizer
Shuyang Hou Haoyue Jiao Ziqi Liu Jianyuan Liang Lutong Xie Xuefeng Guan Zhipeng Gui Huayi Wu a State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing, Wuhan University, Wuhan, Chinab School of Resource and Environmental Sciences, Wuhan University, Wuhan, Chinac School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, ChinaShuyang Hou received his MS degree in Geographic Information Science from Wuhan University and is pursuing a PhD at the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University. His research interests include GeoAI and geospatial code generation. Contributions: Conceptualization, Methodology, Software, Formal analysis, Data curation, Visualization, Writing–original draft, Writing–review & editing.Haoyue Jiao is an undergraduate student in the School of Resource and Environmental Sciences, Wuhan University. Her research interests include map design and geospatial code generation. Contributions: Software, Visualization, Data curation, Validation, Writing–review & editing.Ziqi Liu received his MS degree in Computer Science from China University of Mining and Technology and is pursuing a PhD at the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University. His research interests include GeoAI and automated algorithms. Contributions: Validation, Software, Resources, Data curation, Writing–review & editing.Jianyuan Liang received his MS degree in Geographic Information Science from Eastern Michigan University and is currently a PhD candidate at the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University. His research focuses on geospatial services, geospatial modeling and generative artificial intelligence. Contributions: Validation, Software, Data curation, Writing–review & editing.Lutong Xie received his BS degree from Wuhan University and is pursuing an MS at the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University. His research interests include geospatial code generation and geospatial modeling. Contributions: Validation, Formal analysis, Data curation, Writing–review & editing.Xuefeng Guan received his PhD in Photogrammetry and Remote Sensing from Wuhan University. He is a Professor at the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University. His research interests include high-performance geospatial computing, distributed management and intelligent analysis of spatiotemporal big data. Contributions: Supervision, Project administration, Resources, Writing–review & editing.Zhipeng Gui received his PhD (2011) degrees from Wuhan University. He is a Professor and Vice Dean of the School of Remote Sensing and Information Engineering at Wuhan University. His research focuses on spatiotemporal data mining and geographic information system applications. Contributions: Supervision, Project administration, Resources, Writing–review & editing.Huayi Wu received his MS degree in Probability and Mathematical Statistics and his PhD in Photogrammetry and Remote Sensing from Wuhan University. He is a Professor at the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University. His research interests include high-performance geospatial computing, intelligent geospatial services, geospatial modeling and applications of deep learning and data mining. Contributions: Conceptualization, Supervision, Funding acquisition, Project administration, Writing–review & editing.
2026/08/03
论文
Sustainable Cities and Society
PublisherJournal
比较热浪期间与随后夏季高纬度沿海哥本哈根水体的降温效应
Comparing waterbody cooling during heatwave with subsequent summers in high-latitude coastal Copenhagen

出版日期:2026年10月1日 来源:《可持续城市与社会》(Sustainable Cities and Society),第149卷 作者:Ida Maria Bonnevie

Sustainable Cities and Society
论文
Sustainable Cities and Society
PublisherJournal
面向社会技术韧性的电动汽车充电柔性调控:配电变压器裕量与区位价值分析
Flexing electric vehicle charging for socio-technical resilience: Distribution transformer headroom and locational value analyses

出版日期:2026年10月1日;来源:《可持续城市与社会》(Sustainable Cities and Society),第149卷;作者:Oluwasola O. Ademulegun、Damian Flynn、Bukola Oni、Motasem Bani Mustafa、Adekanmi M. Adeyinka、Eustache Uwimana、Neil J. Hewitt

Sustainable Cities and Society
论文
Sustainable Cities and Society
PublisherJournal
非正式住区室内外热应激、住房特征与健康结局:来自坦桑尼亚达累斯萨拉姆的混合方法研究
Indoor–outdoor heat stress, housing characteristics, and health outcomes in an informal settlement: a mixed‑methods study from Dar es Salaam, Tanzania

出版日期:2026年10月1日;来源:《可持续城市与社会》(Sustainable Cities and Society),第149卷;作者:Tobi Eniolu Morakinyo、Iyanuoluwa Fatunmbi、Lazaro Eliyah Mngumi、Elinorata Mbuya、Francesco Pilla、Oluwafemi Benjamin Obe、David Bassey、Modest Baruti

Sustainable Cities and Society
论文
arXiv
RemoteSensing
EarthObservation
竞争性卫星部署与地球静止轨道的经济地理
Competitive satellite placement and the economic geography of the geostationary orbit

地球静止轨道(GEO)承载了全球大部分卫星通信收入。国际电信联盟(ITU)通过各国主管部门协调并登记频率指配及相应轨道位置,但未就经度位置设定市场价格或产权凭证。本文探讨商业运营商在此行政管理体制下如何在GEO中进行卫星定位。借助空间排序模型,我们发现商业卫星的部署数量与某一GEO经度下方可触达收入成正比,并按空间折现率依观测角予以折现。我们估计的部署弹性接近1,即某一经度上方的卫星数量随其可触达收入同比例增长。相比之下,政府卫星的部署则与某一GEO经度所覆盖的人口规模成正比。我们通过实证验证了该商业模型:基于2012年部署数据估计的参数能较好预测2021年的分布;且该模型拒绝将中国GDP全部视为国际商业卫星舰队的可触达市场。反事实分析表明,低地球轨道(LEO)系统的竞争可能使GEO使用重心向亚太地区倾斜,而更高的空间折现率则可能使GEO使用进一步集中于高GDP国家。

Akhil Rao, Nikodem Szumilo
2026/06/24
论文
arXiv
AI
企业与工人双重迁移驱动的城市形成
City formation by dual migration of firms and workers

本文将新经济地理学中的核心—边缘模型(Core-Periphery model)加以拓展,该模型原本仅考虑由地区间实际工资差异所驱动的工人单向迁移;拓展后新增了由地区间实际利润差异所驱动的企业迁移。在此双重迁移模型中,解的行为在定性上与单迁移模型相似:1)空间上均匀的人口分布变得不稳定,最终形成若干个城市,其中企业与工人共同集聚;2)随着运输成本下降,城市数量减少。这些结果为单迁移模型的应用提供了更一般的理论依据。

Kensuke Ohtake
2023/11/09
论文
Sustainable Cities and Society
PublisherJournal
UrbanComputing
未来城市化与气候变化情景下生物源挥发性有机化合物(BVOC)排放预测:时空大气风险与生态系统服务权衡
Projecting BVOC emissions under future urbanization and climate change: spatiotemporal atmospheric risks and ecosystem service trade-offs

出版日期:2026年10月1日;来源:《可持续城市与社会》(Sustainable Cities and Society),第149卷;作者:董文、王兴宝、周汉星、张悦、黄银成、徐凯、黄东明、刘东、张静、程天亮、陈健、王建武、任远

Sustainable Cities and Society
论文
GIScience & Remote Sensing
PublisherJournal
Platform
超越静态图:一种尺度可变的动态图框架,融合空间与层次关系以实现城市土地利用制图
Beyond static graphs: a scale-variant dynamic graph-based framework integrating spatial and hierarchical relationships for urban land-use mapping
Yu Su Chenguang Dai Yongqi Sun Ruiyi Yang Anzhu Yu Yanfei Zhong a School of Surveying and Mapping, Information Engineering University, P.R. China b State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, P.R. China c Hubei Provincial Engineering Research Center of Natural Resources Remote Sensing Monitoring, Wuhan University, P.R. China Yu Su is a lecturer at Information Engineering University. Her research interests include urban land-use mapping based on multi-source geographic data. She contributed to the conceptualisation, methodology, validation, visualisation, formal analysis, investigation, data curation, writing and funding acquisition. Chenguang Dai is a professor at Information Engineering University. His research interests include remote sensing image interpretation and GIScience. He contributed to the conceptualisation, methodology, formal analysis, investigation, resources, writing and supervision. Yongqi Sun is a student at Information Engineering University. Her research interests include high resolution remote sensing classification and land cover mapping. She contributed to the investigation and data curation. Ruiyi Yang is a student at Wuhan University. Her research interests include remote sensing image scene classification. She contributed to the software and data curation. Anzhu Yu is a professor at Information Engineering University. His research interests include remote sensing image interpretation. He contributed to the conceptualisation, investigation, resources, supervision and funding acquisition. Yanfei Zhong is a professor at Wuhan University. His research interests include remote sensing image interpretation and GIScience. He contributed to the conceptualisation, investigation, resources, supervision and funding acquisition.
2026/08/03
论文
GIScience & Remote Sensing
PublisherJournal
Platform
确保连续高精度地球辐射收支记录:面向风云三号F星ERM-II载荷的场景依赖型光谱融合反滤波框架
Ensuring continuous high-precision Earth Radiation Budget Records: a scene-dependent spectral fusion unfiltering framework for FY-3F ERM-II
Siying Chen Menghui Wang Huizeng Liu Ping Zhu Hong Qiu Jin Qi Wanchun Zhang Xiuqing Hu Tianye Cao Alfateh M. Tag Elsir Qingquan Li a Institute for Advanced Study & Tiandu-Shenzhen University Deep Space Exploration Joint Laboratory & Space Science Center & Institute for Carbon Neutrality, Shenzhen University, Shenzhen, China b MNR Key Laboratory for Geo-Environmental Monitoring of Great Bay Area & Guangdong Key Laboratory of Urban Informatics & Shenzhen Key Laboratory of Spatial Smart Sensing and Services, Shenzhen University, Shenzhen, China c National Satellite Meteorological Center, China Meteorological Administration, Beijing, China
2026/08/03
论文
Sustainable Cities and Society
PublisherJournal
Platform
《面向气候适应性乡村形态:一种基于网格的微网格框架以量化微气候驱动因子》勘误表 [《可持续城市与社会》第148卷(2026年)107617页]
Corrigendum to “Toward climate-adaptive rural morphology: a grid-based micro-mesh framework for quantifying microclimate drivers” [Sustainable Cities and Society 148 (2026) 107617]

出版日期:2026年9月15日;来源:《可持续城市与社会》,第148卷;作者:李志星、苏艳军、王宗贤、田咪咪

Sustainable Cities and Society
论文
arXiv
GeoAI
GIS
TerraNova:面向人类世的基础模型
TerraNova: A Foundation Model for the Anthropocene

人类世的一个核心挑战在于将物理地球系统与人类社会建模为一个耦合系统,然而目前尚无学习所得的表征能够覆盖二者在观测尺度上的全部广度。我们认为其根本障碍在于几何结构差异:物理地球以忽略政治边界的连续场形式被测量,而社会数据则按行政单元(如国家)上报。现有地球系统基础模型适配前一种几何结构;将其与后一种几何结构耦合时,不得不依赖有损的边界平均操作。本文提出 TerraNova——一种在原始几何结构下训练的基础模型,输入包含 1,024 类物理与社会记录:其中 512 类为格网化地球系统场,另 512 类为国家级指标。专用编码器分别表征位置、国家、时间和任务;跨模态 Transformer 将其融合为统一的时空状态;超网络则为每个查询生成专属解码器,其证据性输出头返回预测分布。该表征通过两项对比学习目标实现耦合:一是基于人口加权的国家与其境内地理坐标的对齐;二是与预训练的、承载图像语义的地理空间嵌入对齐。经该解码器读出后,该表征在性能上可媲美专用于地理空间任务的编码器,同时拓展了后者未涵盖的维度(时间、海洋及不确定性),并支持国家级别的建模能力。冻结主干网络可在消费级硬件上数分钟内完成从稀疏观测重建稠密场,并快速适配至未见变量。

Carlos Rodriguez-Pardo, Massimo Tavoni
2026/07/31
论文
arXiv
RemoteSensing
EarthObservation
面向少样本遥感场景分类的局部一致传递式信息最大化方法(LC-TIM)
Locally Consistent Transductive Information Maximization for Few-Shot Remote Sensing Scene Classification

遥感场景分类正日益依赖于在大规模地球观测数据上预训练的基础模型。此外,传递式推理(transductive inference)——即利用整个未标记查询集的整体统计结构——天然契合遥感处理流程,因为在该流程中,大尺寸遥感影像通常被切分为图像块并以批处理方式联合推理。本文提出 LC-TIM(Locally Consistent Transductive Information Maximization),该方法在当前最优的面向少样本 CLIP 的传递式信息最大化(TIM++)目标函数基础上,引入一个局部一致性正则项,强制每个查询样本与其在特征空间中的 $κ$ 个最近邻样本的预测结果保持一致。该正则项以单一乘性因子形式嵌入闭式 $q$-更新中,仅带来可忽略的计算开销。我们进一步提出一种多源扩展方案,融合多个遥感基础模型生成的亲和图(affinity graph),从而进一步提升分类精度。为评估所提方法,我们构建了首个面向传递式少样本遥感场景分类的全面、开源基准,涵盖十个多样化数据集、两种遥感视觉-语言模型以及多种少样本设定,对 LP++、TransCLIP、TIM++ 和 LC-TIM 进行系统评测。实验表明,传递式方法始终优于零样本基线;LC-TIM 达到当前最优分类精度,尤其在低样本量(low-shot)情形下增益最为显著——此时邻域线索最具判别性。代码已开源:https://github.com/elkhouryk/LC-TIM

Karim El Khoury, Benoît Gérin, Benoît Macq
2026/07/31
论文
arXiv
GeoAI
GIS
面向AAM基础设施规划的人本导向噪声烦恼度制图ArcGIS框架
An ArcGIS Framework for Mapping Human-Centered Noise Annoyance for AAM Infrastructure Planning

先进空中交通(Advanced Air Mobility, AAM)代表了城市交通的范式变革;然而,其成功实施高度依赖公众接受度,而低空电动垂直起降(electric vertical takeoff and landing, eVTOL)运行所产生的噪声已成为主要关切。现有研究通常采用A计权声级(LAeq)和昼夜平均声级(Ldn)等声学指标描述eVTOL噪声。本研究构建了一个基于地理信息系统(Geographic Information System, GIS)的框架,将eVTOL声学输出转化为表征高度烦恼人群比例(%HA)的空间分布图,案例为美国阿肯色州西北部一条典型医疗物资配送航线。结果表明,噪声与烦恼度总体随距航线距离增加而递减,其中下降阶段引发的烦恼度最高,其次为爬升阶段和巡航阶段。通过整合美国人口普查(Census)人口数据,估算高度烦恼个体数量并识别空间影响热点区域。进一步将噪声-烦恼度结果与航线距离及空域因素相结合,评估备选航线方案,识别兼顾多方需求的均衡路由策略。该框架将声学评估与人类主观响应相衔接,支持噪声敏感区域识别、航线方案比选以及社会可持续的AAM基础设施规划。

Swapnil Saha, Zubin Mistry, Neelakshi Majumdar
2026/07/31
论文
arXiv
GeoAI
GIS
评估GeoAI解释与遥感领域知识在卫星洪水制图中的对齐程度
Evaluating the Alignment Between GeoAI Explanations and Domain Knowledge in Satellite-Based Flood Mapping

卫星数量的持续增长提升了地球观测的时间分辨率,使基于卫星的洪水制图成为业务化洪水监测的一项有前景的方法。作为地理空间人工智能(GeoAI)的重要应用,基于深度学习的卫星影像洪水制图方法通过从海量遥感数据中学习复杂的空谱模式,展现出更优的预测性能。然而,深度学习模型决策过程的不透明性仍是其融入关键科学与业务工作流的主要障碍。这凸显了系统评估模型解释是否符合既定遥感领域知识的必要性。为填补这一研究空白,本研究提出了ADAGE(领域知识与GeoAI解释对齐评估,Alignment between Domain Knowledge and GeoAI Explanation Evaluation)框架。该框架旨在系统评估深度学习模型解释与既定遥感知识(特别是地表独特光谱特性)之间的对齐程度。ADAGE框架采用通道分组SHAP(SHapley Additive exPlanations)方法,估算分组输入通道对像素级预测的贡献。在两项基于卫星的洪水制图任务上的实验表明,ADAGE框架能够:(1)定量评估模型解释与基于领域知识生成的参考解释之间的对齐程度;(2)借助所提出的对齐分数,辅助领域专家识别未对齐的解释。本研究有助于弥合GeoAI在地球观测中可解释性与领域知识之间的鸿沟,提升其应用可靠性。

Hyunho Lee, Wenwen Li
2026/04/29
论文
International Journal of Digital Earth
PublisherJournal
GeoAI
HiSpot:面向城市分析的空间优化集成且易用的 Python 框架
HiSpot: an integrated and accessible python framework for spatial optimization in Urban Analytics

空间优化在城市分析中发挥着关键作用,支撑从设施选址到物流路径规划等重要决策。然而,这些方法的实际应用往往……

Shaohua Wang Cheng Su Zixuan Zhao Haojian Liang Zhenbo Wang Yeran Sun Yang Zhong Jiayi Zheng Zhuonan Huang Jingyi Zhou a State Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, People's Republic of China b College of Resources and Environment, University of Chinese Academy of Sciences, Beijing, People's Republic of China c Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, People's Republic of China d School of Artificial Intelligence, China University of Geosciences, Beijing, People's Republic of China
2026/07/27
论文
Transactions in GIS
PublisherJournal
GeoAI
利用街景图像估计客观城市感知:以生理唤醒为应用
Estimating Objective Urban Perception by Using Street View Images: An Application to Physiological Arousal

《GIS学报》(Transactions in GIS),2026年8月,第30卷,第5期。

Wei Jiang, Jing Mao, Shuo Gong, Xinyue Zheng, Jingjing Zhao, Yi Long
2026/08/03
论文
Scientific Data
TopJournal
Dataset
全球复杂城市贫困地区的建筑物轮廓人工解译地理数据集
A manually interpreted geo-dataset of building footprints in complex urban poverty areas across the globe

《科学数据》(Scientific Data),在线发表日期:2026年8月3日;doi:10.1038/s41597-026-07945-2。该数据集为全球复杂城市贫困地区的建筑物轮廓人工解译地理数据集。

Nicolas J. Kraff
2026/08/03
论文
arXiv
SpatialIntelligence
Trajectory
FlexComposer:面向图像到动态影像的统一视频合成框架,支持灵活轨迹控制
FlexComposer: Unified Video Compositing from Images to Dynamic Footage with Flexible Trajectory Control

生成式视频合成——即将外部素材无缝嵌入现有视频序列——在内容创作与视觉特效中至关重要。然而,现有方法普遍存在控制性与保真度之间的权衡:或从静态图像中幻化运动,导致预动画素材的动态特性无法保留;或缺乏细粒度空间控制,难以沿用户定义轨迹精确放置素材。我们提出 FlexComposer,一种将视频合成标准化为轨迹引导条件生成任务的统一框架,支持静态图像与动态影像的无缝融合。本方法包含三项核心设计:(1)统一规范前景表示(Unified Canonical Foreground Representation),将物体固有运动与其全局位移解耦,将异构输入标准化为稳定、居中的潜在空间;(2)空间感知潜在特征注入策略(Spatial-Aware Latent Injection),利用变分自编码器(VAE)潜在空间的平移等变性,通过无参数机制将规范特征映射至目标轨迹;(3)混合数据集与合成到真实渐进式训练范式(Hybrid Dataset and Synthetic-to-Real Curriculum),协同整合程序化仿真、真实电影镜头及生成数据,隐式学习符合物理规律的光照与阴影协调。该统一设计可处理从产品照片到动态主体的多样化输入,在无需显式三维重建或辅助可学习适配器的前提下,实现高保真运动控制与环境融合。大量实验表明,FlexComposer 在视觉质量、时序一致性与轨迹贴合度方面均优于当前最优方法。

Songchun Zhang, Sitong Guo, Xianghao Kong
2026/08/01
论文
Sustainable Cities and Society
PublisherJournal
UrbanComputing
北京市中心城区PM2.5暴露的长期动态变化及模拟健康收益:多维环境条件下时空分异与分层约束
Long-term dynamics of PM 2.5 exposure and modeled health gains in Beijing’s central urban area: Spatiotemporal differentiation and layered constraints across multidimensional environmental conditions

出版日期:2026年10月1日;来源:《可持续城市与社会》(Sustainable Cities and Society),第149卷;作者:刘玉山、赵静

Sustainable Cities and Society
论文
arXiv
RemoteSensing
EarthObservation
贫困制图:数据、模型与应用
Poverty Mapping: Data, Models and Applications

贫困制图对于监测联合国2030年可持续发展议程中的可持续发展目标1(SDG 1)——即在全球各地消除一切形式的贫困——日益重要。然而,及时且高空间分辨率的贫困估计仍面临困难,因为传统的基于人口普查和调查的方法成本高昂、实施频次低,且在贫困最严重的地区往往数据最为稀疏。由于贫困源于由人类流动性、社会互动、基础设施和经济活动共同塑造的复杂社会经济系统,新兴计算方法与非传统数据源为贫困估计与制图带来了新机遇。这些方法处于统计物理、复杂系统科学与数据科学的交叉领域,能够实现更高时空分辨率的贫困估计。本文综述了贫困的基本概念及主要测度框架,并考察了利用卫星影像、手机数据、社交媒体数据以及多源数据融合进行贫困估计与制图的最新进展。本文还探讨了代表性、跨区域可迁移性、可解释性及不确定性量化等长期存在的挑战。最后,本文厘清了当代贫困制图在分析潜力与实际应用限制两方面的边界。

Suoyi Tan, Mengning Wang, Yixiu Kong
2026/07/31
论文
arXiv
SpatialIntelligence
Trajectory
源中心状态演化循环 Transformer
Looped Transformers with Source-Centered State Evolution

循环 Transformer(Looped Transformers)通过在递归深度上复用同一 Transformer 模块,构建了一条对训练与推理均有效的计算轴,在参数量固定的前提下提升了有效深度。然而,该共享模块必须调控整个隐藏状态轨迹——该轨迹在训练及外推深度下均呈现动态变化。此外,在加性注入式循环 Transformer 中,输入条件信号在每一递归步均被重新引入,因此即使在输入条件参考点处应用共享状态转移,仍可能导致隐藏状态发生偏移。本文提出源中心状态演化(Source-Centered State Evolution, SCSE),旨在协调输入条件性与参考点保持型共享递归。具体而言,SCSE 通过其可学习锚点(anchor)与初始偏差(deviation)维持输入依赖性;允许非零偏差驱动递归计算,同时将零偏差映射为零;并通过零偏差掩码(zero-deviation mask)确保锚点的精确不变性。由此,指定锚点在构造上即为一步不动点(one-step fixed point)。零偏差强制偏置(zero-deviation forcing bias)指从锚点自身生成的下一偏差,在 SCSE 中该偏置消失,而非零偏差则保持活跃并支持状态依赖的递归计算。理论分析表明,零偏差强制偏置是一种设计自由度,其任务效应可能有害、中性或有益;SCSE 通过将其设为零,以实现锚点的精确不变性,从而消解该设计选择问题。在 WikiText-2、WikiText-103、直接网络语料预训练、保留网络文本迁移任务以及 LAMBADA 完形填空任务上,SCSE 均提升了受控递归质量前沿。消融实验表明,可学习锚点与锚点坐标系下的偏差递归是性能增益的主要来源;一项针对训练模型的案例研究进一步验证了上述结论。

Bum Jun Kim, Kohei Hayashi, Shunsuke Kamiya
2026/07/30
论文
arXiv
Trajectory
Mobility
A2TTA:面向演进型交通传感器网络的锚定-敏捷测试时自适应方法
A2TTA: Anchored-and-Agile Test-Time Adaptation for Evolving Traffic Sensor Networks

交通预测对智慧城市中的高效交通管理和路径规划至关重要。现有交通预测研究通常假设传感器图结构固定,忽视了现实世界交通网络的持续演化,例如道路网络的持续建设与人类移动模式的动态变化。此类动态变化会显著降低传统预测模型的性能,从而促使研究者采用测试时自适应(Test-Time Adaptation, TTA)方法,在部署阶段高效调整预训练模型。然而,将TTA应用于演进型交通传感器网络仍面临两大挑战:其一,拓扑扩展引入新传感器及连接关系,持续重塑传感器图;其二,时间偏移在时间尺度与稳定性上存在差异,需对长期与短期偏移实施差异化自适应。本研究提出A2TTA——一种面向演进型交通传感器网络的锚定-敏捷测试时自适应框架,该框架将拓扑演化引发的预测误差转化为可扩展的输出校准问题,并将时间自适应解耦为持久性全局校正与敏捷的上下文特异性专业化。通过协同应对拓扑演化与多尺度时间偏移,A2TTA实现了对持续演化的交通环境的高效且鲁棒的自适应。在十个真实交通网络上的大量实验表明,A2TTA在不同骨干网络、数据集及预测时域下均能持续提升预测性能。代码开源地址:https://github.com/lixus7/A2TTA。

Du Yin, Xiachong Lin, Yue Tan
2026/07/28
论文
arXiv
SpatialIntelligence
UrbanTraffic
SqLinear:均衡的方形划分使线性交互足以支撑大规模交通预测
SqLinear: Balanced Square Partitioning Makes Linear Interaction Sufficient for Large-Scale Traffic Forecasting

交通预测是智能交通系统与城市尺度决策的核心任务。尽管主流基于神经网络的方法效果显著,但其在部署于拥有数千个交通传感器的真实场景时,因计算可扩展性差而严重受限。为解决此问题,研究社区尝试引入空间数据库划分技术以提升模型可扩展性。然而,这些方法依赖人工设计的几何启发式策略,常导致不规则或不平衡的数据划分,进而引发边界碎片化、填充开销过大及模型精度下降等问题。本文提出 SqLinear,一种面向大规模交通预测的高效且有效的架构。首先,我们设计了方形划分(Square Partition),一种几何自适应算法,将海量交通传感器划分为均衡、互不重叠且空间紧凑的区域。与现有基于启发式的设计不同,方形划分具有理论基础,可对划分利用率与分割平衡性提供可证明的保障,从而为下游时空建模奠定高质量基础。其次,我们提出分层线性交互(Hierarchical Linear Interaction, HLI)模块,摒弃了 Transformer 类时空模型中常见的高成本注意力机制。HLI 通过轻量级线性交互方案高效传播全局区域间依赖关系,并在节点层级对其进行细化,从而以线性计算复杂度实现有效的时空建模。在四个大规模交通数据集及 11 种基线方法上的大量实验表明,SqLinear 在标准设置下平均降低 MAE 2.30%,在极端可扩展性设置下最高降低 MAE 6.78%,同时减少……

Yongfeng Su, Hongwen Li, Zijian Zhang
2026/06/19
1 / 256