SHIELD: using foundational vision encoders for real-time activity detection in satellite imagery

SHIELD (Strategic Humanitarian Insights for Emergency Logistics and Distribution) is an automated pipeline that converts satellite imagery into actionable, near-real-time maps of human-activity change. For each Area of Interest (AOI), SHIELD tiles imagery encodes tiles with vision–language model embeddings and projects them to a compact latent space. A spatial mixture density network parameterizes a per-tile Gaussian-mixture baseline of “pattern-of-life,” against which new observations are scored by log-likelihood. Statistically significant outliers are filtered and aggregated into AOI heatmaps, which are delivered via a lightweight web dashboard within hours of image availability. We evaluate SHIELD on two scenarios: (i) rapid tent-city expansion near Kibati, Democratic Republic of Congo, and (ii) wildfire burn-scar detection and recovery dynamics in Los Angeles, California (January 2025). Across both, SHIELD reduces spurious alarms relative to spectral differencing while preserving sensitivity to semantically informed change. By combining modern vision encoders with probabilistic baselines, SHIELD advances remote-sensing change detection toward explainable, scalable, and operational situational awareness for humanitarian response.

• Publication year

  2025

• Venue

  Future Sensing Technologies

• Publication date

  2025-12-15

• Fields of study

  Computer Science, Engineering, Environmental Science

• Identifiers
  DOI 10.1117/12.3074932
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Situational Awareness: Techniques, Challenges, and Prospects

  2022cited by this paper
• Learning Transferable Visual Models From Natural Language Supervision

  2021cited by this paper
• A Novel Framework for the Design of Change-Detection Systems for Very-High-Resolution Remote Sensing Images

  2013cited by this paper
• Image change detection algorithms: a systematic survey

  2005cited by this paper
• Map Projections

  1947cited by this paper

• No citing papers are available for this paper.