用深度学习融合人口普查数据，提升城市形态大尺度空间分解的精度与可解释性。
DeepC4: Deep Conditional Census-Constrained Clustering for Large-scale Multitask Spatial Disaggregation of Urban Morphology
- 引入条件约束聚类，将真实人口普查数据作为聚类层级的监督信号。
- 在卢旺达实验中，房屋与居民数量估算误差低于1.2%，优于现有方法。
- 适合需要高精度、可审计的城市规划与灾害风险评估研究者使用。
为理解全球可持续发展及发展中经济体灾害风险减缓进展，全球地震模型（GEM）基金会与地球观测暴露建模项目（METEOR）采用经典空间分解技术，利用卫星影像、地理空间数据与分省人口普查信息生成大尺度城市形态地图。然而，局部与已验证人口普查数据的偏差及模型不确定性传播仍是粗粒度到细粒度映射的核心挑战，尤其受限于弱且条件化的标签监督。为此，本文提出深度条件人口普查约束聚类（DeepC4），一种基于深度学习的空间分解方法，将局部人口普查数据作为聚类级约束，并在多任务学习框架下联合建模卫星影像模式与多重条件标签关系。以卢旺达城市形态为例，DeepC4在屋顶、墙体、高度预测上的宏F1分别为0.63、0.78、0.45，宏mIoU分别为0.57、0.71、0.42；全国房屋与居民数量估计误差分别仅为1.13%和1.11%，优于GEM（2.03%和3.29%），且覆盖的500米网格像素数比METEOR多32%-49%。随着2030年多项全球框架临近，本工作提供了一种显式编码真实人口普查与专家知识的深度学习映射技术，实现对大尺度粗粒度信息的可解释性审计。
原文摘要 · Abstract (English)
To understand our global progress for sustainable development and disaster risk reduction in many developing economies, two recent major initiatives - the Uniform African Exposure Dataset of the Global Earthquake Model (GEM) Foundation and the Modelling Exposure through Earth Observation Routines (METEOR) Project - implemented classical spatial disaggregation techniques to generate large-scale mapping of urban morphology using the information from various satellite imagery and its derivatives, geospatial datasets of the built environment, and subnational census statistics. However, the local discrepancy with well-validated census statistics and the propagated model uncertainties remain a challenge in such coarse-to-fine-grained mapping problems, specifically constrained by weak and conditional label supervision. Therefore, we present Deep Conditional Census-Constrained Clustering (DeepC4), a novel deep learning-based spatial disaggregation approach that incorporates local census statistics as cluster-level constraints while considering multiple conditional label relationships in a joint multitask learning of the patterns of satellite imagery. As a demonstration using Rwandan urban morphology, DeepC4 achieves macro-F1 scores of 0.63, 0.78, and 0.45 and macro-mIoU of 0.57, 0.71, and 0.42 for roof, wall, and height prediction respectively, estimates national dwelling and occupant counts within 1.13% and 1.11% error compared to census records, outperforming GEM (2.03% and 3.29%), and occupies 32%-49% more 500-meter grid pixels than METEOR across provinces. As the world approaches the conclusion of many global frameworks in 2030, our work offers a new deep learning-based mapping technique that explicitly encodes well-validated census and experts' belief systems to achieve an explainable and interpretable auditing of existing coarse-grained derived information at large scales.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。