Semantic Segmentation for Sequential Historical Maps by Learning from Only One Map

TL;DR

The paper addresses digitizing historical maps by introducing a weakly supervised semantic segmentation framework that leverages temporal consistency through age-tracing. By pre-training on a single labeled anchor year and progressively fine-tuning across neighboring years using pseudo-labels, the approach achieves strong cross-temporal generalization on the Hameln dataset, with mean IoU up to 77.3% and overall accuracy around 97%. Key contributions include the age-tracing strategy, the creation of the Hameln dataset, and empirical evidence that bi-directional tracing outperforms baselines while carefully managing pseudo-label uncertainty. This method enables scalable digitization of historical maps with substantially reduced labeling effort and has practical implications for long-term land-use and urban-change studies.

Abstract

Historical maps are valuable resources that capture detailed geographical information from the past. However, these maps are typically available in printed formats, which are not conducive to modern computer-based analyses. Digitizing these maps into a machine-readable format enables efficient computational analysis. In this paper, we propose an automated approach to digitization using deep-learning-based semantic segmentation, which assigns a semantic label to each pixel in scanned historical maps. A key challenge in this process is the lack of ground-truth annotations required for training deep neural networks, as manual labeling is time-consuming and labor-intensive. To address this issue, we introduce a weakly-supervised age-tracing strategy for model fine-tuning. This approach exploits the similarity in appearance and land-use patterns between historical maps from neighboring time periods to guide the training process. Specifically, model predictions for one map are utilized as pseudo-labels for training on maps from adjacent time periods. Experiments conducted on our newly curated \textit{Hameln} dataset demonstrate that the proposed age-tracing strategy significantly enhances segmentation performance compared to baseline models. In the best-case scenario, the mean Intersection over Union (mIoU) achieved 77.3\%, reflecting an improvement of approximately 20\% over baseline methods. Additionally, the fine-tuned model achieved an average overall accuracy of 97\%, highlighting the effectiveness of our approach for digitizing historical maps.

Figures (6)

• Figure 1: An example of historical maps and the corresponding labels of Hameln, Germany.
• Figure 2: Pipeline for training the UNet by age-tracing.
• Figure 3: Comparison of labels from 1974 (green) and 2023 (red). The overlapping areas (yellow) are the consistent labels. The percentage values in the brackets are the IoUs between the labels from the two ages.
• Figure 4: Exemplar result of mono-directional age-tracing for class FW, SW, SM.
• Figure 5: Semantic segmentation performance with different uncertainty thresholds for pseudo labels of bi-directional age-tracing.