EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Mapping the Asymmetric Spatial Distribution of Urban Homeless Encampments Using High-Resolution Satellite Imagery

Jiale Qian

IEEE Access · 2026 · 14 · 48859-48871

DOI: 10.1109/access.2026.3678098 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Jiale Qian. Die Arbeit nennt 74 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Homeless encampments are a visible manifestation of the housing crisis in many major cities. Accurate spatial mapping of these encampments is critical for urban planning, resource allocation, and public health responses, yet it remains methodologically challenging due to the visual ambiguity of encampment features and the scarcity of reliable training data. This study presents an integrated, end-to-end methodological framework for detecting and mapping homeless encampments from high-resolution satellite imagery, applied to the Skid Row neighborhood of Los Angeles. We address annotation uncertainty through the application of the Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm to resolve inter-annotator disagreement and produce a probabilistic consensus ground-truth dataset. We then benchmark four state-of-the-art instance segmentation models—Mask R-CNN, Cascade Mask R-CNN, YOLACT, and YOLO v8—under a rigorous 10-fold cross-validation framework. YOLO v8 achieves the best performance across all metrics, attaining a segmentation mAP of 0.583 and a mAP50 of 0.840, substantially outperforming the next-best model (Cascade Mask R-CNN: seg mAP = 0.486). Ablation experiments confirm that this advantage is attributable to two specific architectural components—the C2f feature extraction module and the multi-scale PAN-FPN fusion mechanism—with the latter identified as the more critical driver of performance on small and irregular targets (Seg mAP decline of 0.096 upon removal versus 0.042 for the C2f module). A heuristic geospatial post-processing filter integrating building footprint and sidewalk proximity data is then applied to suppress false positives; this step reduces the false positive count from 249 to 26 while leaving true positives unchanged, improving precision from 78.3% to 97.2% and the F1 score from 86.7% to 97.2%. The final predictions are used to generate a GIS-ready spatial database of encampments. This framework provides a scalable and reproducible approach to encampment mapping that can support evidence-based urban policy and resource deployment.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Qian, J. (2026). Mapping the Asymmetric Spatial Distribution of Urban Homeless Encampments Using High-Resolution Satellite Imagery. IEEE Access, 14, 48859-48871. https://doi.org/10.1109/access.2026.3678098

Vancouver

Qian J. Mapping the Asymmetric Spatial Distribution of Urban Homeless Encampments Using High-Resolution Satellite Imagery. IEEE Access. 2026;14:48859-48871. doi:10.1109/access.2026.3678098

BibTeX
@article{Qian2026,
  author = {Qian, Jiale},
  title = {Mapping the Asymmetric Spatial Distribution of Urban Homeless Encampments Using High-Resolution Satellite Imagery},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {48859--48871},
  doi = {10.1109/access.2026.3678098}
}
Kostenlos starten →