A two-stage contour optimization network for building outline extraction from remote sensing images
Building extraction from remote sensing images is the basis of many applications. However, most current studies consider building extraction a semantic segmentation task to categorize pixels in an image. In this paper, a two-stage building object extraction model is constructed: contour initializati...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-12-01
|
| Series: | Geocarto International |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/10106049.2025.2454940 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Building extraction from remote sensing images is the basis of many applications. However, most current studies consider building extraction a semantic segmentation task to categorize pixels in an image. In this paper, a two-stage building object extraction model is constructed: contour initialization and contour optimization. We directly extract and predict the location of each contour vertex of the building outline by designed model. It mitigates matching errors between contour vertices by dividing the contour into multiple segments. Moreover, an edge attention module is introduced in the contour optimization stage, which adjusts the weight of each vertex in the regression process. So that, vertices far from the real contour are forced to move more quickly. On the Vaihingen dataset, our method achieved 96.67% and 95.95% scores in the Wconv and F1 scores. For the SpaceNet dataset, our proposed method also achieved a very competitive F1 score. |
|---|---|
| ISSN: | 1010-6049 1752-0762 |