Our GeoAI methodology and deep learning processes have demonstrated high accuracy in computer vision. More significantly, they enabled us to analyze historical aerial and satellite imagery alongside historical demographic data from an interdisciplinary viewpoint, integrating remote sensing, population geography, and geospatial data science. A major outcome of this project is the scalability of our methodology, showcased by the semantic segmentation and public sharing of a benchmark land cover dataset for KeyHole-9 Hexagon reconnaissance satellite imagery from the 1970s and the 1980s, which provides nearly global high-resolution coverage. This Hexagon imagery was classified and made accessible online as recently as 2020 and 2022. As a result, we believe our methodology is both valuable and timely for future research.
In addition, our benchmark HexaLCSeg dataset (Hexagon-based Historical Land Cover Benchmark Dataset,
https://zenodo.org/records/11005344(odnośnik otworzy się w nowym oknie)) derived from Hexagon satellite imagery, offers significant potential for various applications. It is particularly useful for environmental and agricultural monitoring and urban planning, as it provides insights into historical land cover conditions and long-term changes such as urbanization, deforestation, and land abandonment. The historical perspective provided by the HexaLCSeg dataset makes it an essential tool for climate change research, allowing the tracking of ecological changes over decades. Furthermore, from the standpoint of geohumanities and human geography, the spatiotemporal characteristics of Hexagon imagery, including urban sprawl in Türkiye and significant rural depopulation in Bulgaria, are invaluable for analyzing and modeling the major dynamics of past and future population geography.
