DeepGolf: A fine-grained perception framework for golf course distribution in the real world based on multi-source remote sensing data
Golf courses, while primarily serving as recreational spaces for high-income populations, occupy significant land areas and thus require precise spatial mapping to support land use planning and environmental management. Traditionally, it has been prohibitively expensive to accurately measure their b...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-02-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S156984322500041X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Golf courses, while primarily serving as recreational spaces for high-income populations, occupy significant land areas and thus require precise spatial mapping to support land use planning and environmental management. Traditionally, it has been prohibitively expensive to accurately measure their built-up areas. This paper presents DeepGolf, an advanced framework that integrates geographic information with deep neural features to accurately extract golf course distributions in real-world scenarios. DeepGolf is structured into three main modules: geographic query, detection, and fine extraction. It employs a robust detection strategy that integrates surface cover, wind speed, and settlement data to identify candidate regions, which is then followed by a grid-based recognition approach and a lightweight Dual-Path Difference Fusion Enhancement (DPDFE) network for detailed localization and fine extraction. When applied across Japan, DeepGolf successfully identifies 2,113 golf courses, covering an area of approximately 2,273 km2 (∼0.6 % of Japan’s land area). The geographic query and detection modules reduce candidate areas by 95.5 % and 98.9 %, respectively. The DPDFE network further enhances accuracy, improving IoU and F1 scores by 2.1 % and 1.2 %, achieving an overall accuracy of 89.8 %. Results indicate that golf courses are primarily concentrated around urban fringes, with the highest densities in the Kanto and Kinki regions. Non-urban areas contain 97.3 % of the courses, while urban regions account for just 2.7 %. DeepGolf serves as an effective tool for precise golf course mapping, supporting enhanced environmental management and resource planning. |
|---|---|
| ISSN: | 1569-8432 |