Accurate building segmentation using multi-source imagery: Integrating satellite data and multi views, multi heights drone images
|
Full Text |
Pdf
|
|
Author |
Akram H. Jalil and Imzahim A. Alwan
|
|
e-ISSN |
1819-6608 |
|
On Pages
|
104-116
|
|
Volume No. |
21
|
|
Issue No. |
2
|
|
Issue Date |
March 20, 2026
|
|
DOI |
https://doi.org/10.59018/012621
|
|
Keywords |
segmentation, deep learning, pre-trained models, vision transformer, self-attention, multi-views, multi-height, drone images.
|
Abstract
Building image segmentation is a critical task in urban planning, disaster management, and environmental monitoring. Traditional segmentation methods using either satellite or drone imagery face challenges such as resolution limitations and vegetation occlusion. In this study, a novel approach integrating multi-source remote sensing data-combining drone imagery captured at multiple heights (150m, 200m, 250m, 300m) with multi-view perspectives combined with satellite imagery to enhance building segmentation accuracy. A key challenge in segmentation is the interference of vegetation and shadows, which can obscure building boundaries. To address this, advanced vegetation removal techniques have been incorporated to refine the extracted structures. By leveraging the complementary advantages of drone and satellite imagery, this approach improves segmentation robustness and reliability. Our method achieves a segmentation accuracy of 96%, significantly outperforming conventional techniques. This approach has been evaluated against existing segmentation methods, demonstrating its effectiveness in extracting high-precision building footprints, even in complex environments. The findings highlight the potential of integrating multi-source data and vegetation suppression strategies to enhance automated urban feature mapping.
Back