Leveraging Segment Anything Model in Identifying Buildings within Refugee Camps (SAM4Refugee) from Satellite Imagery for Humanitarian Operations

TL;DR

The paper addresses efficient building footprint extraction in refugee camps from satellite imagery under limited labeled data. It adopts SAM-Adapter to adapt a foundation model for semantic segmentation, demonstrating superior accuracy across five camp sites and two data sizes, outperforming traditional and Transformer baselines. A key finding is that upscaling with super-resolution, particularly EDSR, markedly improves segmentation quality and can yield peak performance in the first training epoch. The work also delivers an open-source, end-to-end workflow for data preparation, model training, inference, and geospatial output in Shapefiles, enabling rapid humanitarian deployment.

Abstract

Updated building footprints with refugee camps from high-resolution satellite imagery can support related humanitarian operations. This study explores the utilization of the "Segment Anything Model" (SAM) and one of its branches, SAM-Adapter, for semantic segmentation tasks in the building extraction from satellite imagery. SAM-Adapter is a lightweight adaptation of the SAM and emerges as a powerful tool for this extraction task across diverse refugee camps. Our research proves that SAM-Adapter excels in scenarios where data availability is limited compared to other classic (e.g., U-Net) or advanced semantic segmentation models (e.g., Transformer). Furthermore, the impact of upscaling techniques on model performance is highlighted, with methods like super-resolution (SR) models proving invaluable for improving model performance. Additionally, the study unveils intriguing phenomena, including the model's rapid convergence in the first training epoch when using upscaled image data for training, suggesting opportunities for future research. The codes covering each step from data preparation, model training, model inferencing, and the generation of Shapefiles for predicted masks are available on a GitHub repository to benefit the extended scientific community and humanitarian operations.

Figures (16)

• Figure 1: A general workflow of data sampling and preprocessing.
• Figure 2: Visual Comparison of Upscaling Methods: Bilinear and Nearest Interpolation and EDSR for Minawao, Dagahaley and Djibo refugee camps.
• Figure 3: Brief architecture of SAM-Adapter.
• Figure 4: Qualitative results of predicted masks from SAM-Adapter, SAM without fine-tuning, and FPN-MiT in Kutupalong refugee camp.
• Figure 5: Qualitative results of predicted masks from SAM-Adapter, SAM without fine-tuning, and FPN-MiT in Nduta refugee camp.