[論文レビュー] Deep learning enables urban change profiling through alignment of historical maps

Sidi Wu, Yizi Chen+5 more|arXiv (Cornell University)|Feb 2, 2026

Human Mobility and Location-Based Analysis被引用数 0

ひとこと要約

The paper presents a fully automated, deep learning framework that aligns historical maps, extracts urban instances, and profiles fine-grained urban change, demonstrated on Paris 1868–1937.

ABSTRACT

Prior to modern Earth observation technologies, historical maps provide a unique record of long-term urban transformation and offer a lens on the evolving identity of cities. However, extracting consistent and fine-grained change information from historical map series remains challenging due to spatial misalignment, cartographic variation, and degrading document quality, limiting most analyses to small-scale or qualitative approaches. We propose a fully automated, deep learning-based framework for fine-grained urban change analysis from large collections of historical maps, built on a modular design that integrates dense map alignment, multi-temporal object detection, and change profiling. This framework shifts the analysis of historical maps from ad hoc visual comparison toward systematic, quantitative characterization of urban change. Experiments demonstrate the robust performance of the proposed alignment and object detection methods. Applied to Paris between 1868 and 1937, the framework reveals the spatial and temporal heterogeneity in urban transformation, highlighting its relevance for research in the social sciences and humanities. The modular design of our framework further supports adaptation to diverse cartographic contexts and downstream applications.

研究の動機と目的

• Motivate the need to quantify long-term urban change from historical maps beyond manual inspection.
• Develop a modular pipeline combining dense map alignment, multi-temporal object detection, and change profiling.
• Enable analysis that is robust to cartographic variation, misalignment, and document degradation.

提案手法

• Dense displacement field estimation for pixel-level map alignment guided by cartographic knowledge through self-synthesis and cycle-consistency training.
• Multi-temporal feature fusion within a shared encoder–decoder to extract urban instances across years.
• Instance segmentation using Mask2Former enhanced with temporal information from multiple map editions.
• Aggregation of instance changes into building blocks to derive block-level change profiles.
• Quantification of change via IoU-based block-level indicators and analysis of spatial-temporal patterns.

Figure 1 : Historical maps violate assumptions underlying standard change-detection and image-alignment methods . ( a ) Two satellite images [ 10 ] with a time difference of 5-10 years exhibit stable object locations across years, allowing simple overlays to reveal changes. ( b ) Historical maps fro

実験結果

リサーチクエスチョン

• RQ1 How can historical map misalignments and cartographic variations be mitigated to enable reliable fine-grained change analysis across decades?
• RQ2 Can a deep learning pipeline jointly align historical maps, extract robust building instances, and produce quantitative change profiles suitable for urban history research?
• RQ3 What are the spatial and temporal patterns of urban transformation in Paris from 1868 to 1937 when using automated, instance-based change metrics?

主な発見

Image	Object	Disp.	Temp. Cons.	SSIM	CD(10%)	CD(20%)	CD(40%)	mV	L1
Unaligned	Training Area	WarpC	GLU-Net	0.772	2.57	1.73	0.82	0.66	7.16
Unaligned	Training Area	Ours	(Ours)	0.830	1.44	0.93	0.43	0.23	2.83
Unaligned	Testing Area	Unaligned	GLU-Net	0.800	3.29	1.54	0.31	-	-
Unaligned	Testing Area	WarpC	(WarpC)	0.880	1.20	0.80	0.35	0.23	2.46
Unaligned	Testing Area	Ours	(Ours)	0.880	1.19	0.79	0.34	0.19	1.73

• The proposed framework achieves strong alignment performance with comparable SSIM to baselines and smoother, more temporally consistent displacement fields.
• Multi-temporal fusion improves instance detection accuracy, especially for medium-sized buildings, and reduces variability across years.
• Instance-based change analysis at the building-block level reveals heterogeneous, geographically clustered urban transformation in Paris, with intense change during 1868–1878 and pronounced outskirts restructuring.
• Building blocks provide coherent, interpretable units for analyzing street-network evolution and infrastructure-focused changes.
• Label propagation and temporal fusion reduce manual annotation effort, enabling scalable analysis across large historical map collections.

Figure 2 : Method overview. ( a ) Workflow for historical map alignment and change analysis. Maps covering the same location are first rectified to a common image plane. For map pairs from different editions (years), we estimate a dense displacement field for pixel-level alignment and extract urban