.png){kind=icon}
.png){kind=icon}
Return
Map-Guided Cross-Training for Building Detection
.png){kind=icon}
PRE
AI
DOI：10.1109/LGRS.2025.3650383.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Extraction of building footprints from freely available, high temporal resolution satellite imagery is important for applications ranging from disaster response to monitoring informal settlements. While Sentinel-2 (S2) is globally available, limited spatial resolution restricts the reliable detection of small or densely packed structures. We propose a novel framework that retargets a super-resolution backbone enhanced super resolution generative adversarial network (ESRGAN) to generate cartographic map-style outputs directly and fuses it with a semantic segmentation network through joint training. Unlike pipelines treating super-resolution (SR) and detection as sequential tasks, our approach leverages map-centric supervision, unifying enhancement and building delineation. For evaluation, we introduce the S2-BDOT-PL dataset, integrating S2 imagery with OpenStreetMap and BDOT10k ground truth. Results demonstrate improved precision and robustness in small-building detection compared to native resolution baselines.
Keywords:
Superresolution
Buildings
Training
Accuracy
Cartography
Spatial resolution
Semantic segmentation
Pipelines
Satellites
Vectors
Remote sensing
satellite mapping of buildings
semantic segmentation
Sentinel-2 (S2)
super-resolution (SR)
Journal
I
IF:
4.4
Papers:
579
Citations:
0