Research Project: GeoAI_LULC_Seg
Loading...
Contributors
Funders
ID
EC.00175
Authors
Kabadayı, Mustafa Erdem
Faculty Member
Publications
Automatic road extraction from historical maps using transformer-based SegFormers
(Multidisciplinary Digital Publishing Institute (Multidisciplinary Digital Publishing Institute (MDPI)), 2024) Kabadayı, Mustafa Erdem; Sertel E., Hucko C.M.; Department of History; Yes; College of Social Sciences and Humanities
Historical maps are valuable sources of geospatial data for various geography-related applications, providing insightful information about historical land use, transportation infrastructure, and settlements. While transformer-based segmentation methods have been widely applied to image segmentation tasks, they have mostly focused on satellite images. There is a growing need to explore transformer-based approaches for geospatial object extraction from historical maps, given their superior performance over traditional convolutional neural network (CNN)-based architectures. In this research, we aim to automatically extract five different road types from historical maps, using a road dataset digitized from the scanned Deutsche Heereskarte 1:200,000 Türkei (DHK 200 Turkey) maps. We applied the variants of the transformer-based SegFormer model and evaluated the effects of different encoders, batch sizes, loss functions, optimizers, and augmentation techniques on road extraction performance. Our best results, with an intersection over union (IoU) of 0.5411 and an F1 score of 0.7017, were achieved using the SegFormer-B2 model, the Adam optimizer, and the focal loss function. All SegFormer-based experiments outperformed previously reported CNN-based segmentation models on the same dataset. In general, increasing the batch size and using larger SegFormer variants (from B0 to B2) resulted in improved accuracy metrics. Additionally, the choice of augmentation techniques significantly influenced the outcomes. Our results demonstrate that SegFormer models substantially enhance true positive predictions and resulted in higher precision metric values. These findings suggest that the output weights could be directly applied to transfer learning for similar historical maps and the inference of additional DHK maps, while offering a promising architecture for future road extraction studies. © 2024 by the authors.
HexaLCSeg: A historical benchmark dataset from Hexagon satellite images for land cover segmentation
(Institute of Electrical and Electronics Engineers Inc., 2024) Kabadayı, Mustafa Erdem; Şengül, Gafur Semi; Tümer, İlay Nur; Sertel, Elif; Department of History; VPRI (Vice Presidency for Research and Innovation); Yes; Administrative Unit; College of Social Sciences and Humanities
Historical land cover (LC) maps are significant geospatial data sources used to understand past land characteristics and accurately determine the long-term land changes that provide valuable insights into the interactions between human activities and the environment over time. This article introduces a novel open LC benchmark dataset generated from very high spatial resolution historical Hexagon (KH-9) reconnaissance satellite images to be used in deep learning (DL)-based image segmentation tasks. This new benchmark dataset, which includes very high-resolution (VHR) mono-band Hexagon images of several Turkish and Bulgarian territories from the 1970s and 1980s, covers a large geographic area. Our dataset includes eight LC classes inspired by the European Space Agency (ESA) WorldCover project except for the tree class, which we divided into subclasses, namely agricultural fruit trees and other trees. We implemented widely used U-Net++ and DeepLabv3+ segmentation architectures with appropriate hyperparameters and backbone structures to demonstrate the versatility and impact of our HexaLCSeg dataset and to compare the performance of these models for accurate and fast LC mapping of past terrain conditions. We achieved the highest accuracy using U-Net++ with an SE-ResNeXt50 backbone and obtained an F1-score of 0.8804. The findings of this study can be applied to different geographical regions with similar Hexagon images, providing valuable contributions to the field of remote sensing and LC mapping. Our dataset, related source codes, and pretrained models are available at https://github.com/RSandAI/HexaLCSeg and https://doi.org/10.5281/zenodo.11005344.
Revealing urban-rural migration corridors through edge-path bundling of historical census data in Bulgaria
(Sage Publishing, 2025) Kabadayı, Mustafa Erdem; Gerrits, Petrus J.; Solomon, Guy; Basiri, Ana; Department of History; Yes; College of Social Sciences and Humanities
This featured graphic visualises internal migration flows in Bulgaria using settlement-level census data from 1934 to 1946, highlighting how topography and local factors shaped migration and urbanisation. By combining spatial interaction models (SIMs) and edge-path bundling techniques, we reveal fine-scale migration corridors and the role of mountainous terrain in influencing mobility. The SIM approach goes beyond aggregate district analysis to provide new insights into rural depopulation and urban growth.
