Earth-observation archives are expanding exponentially and increasingly contain multi-temporal, multi-spectral, and multi-sensor imagery. Conventional catalogues are effective for searching by metadata such as coordinates, acquisition date, platform, and product type, but analysts must know where and when to look before examining imagery. Translating foundation models into reliable on-premises operational systems remains challenging, especially when ingesting new acquisitions incrementally, preserving geospatial provenance, and suppressing false change caused by seasons, atmosphere, or sensor angles.
Develop a system that makes satellite-imagery archives queryable by semantic content and by multi-temporal change over time, while retaining conventional spatial, temporal, and sensor filters. Support analyst discovery through natural language queries and image-to-image similarity search.
Translating vision-language foundation models into low-latency on-premises vector indexes, handling petabyte-scale raster tiling, and isolating true tactical structural change from environmental noise like seasonal phenology and sun-angle shifts.
A multimodal Earth-Observation retrieval system supporting natural language search (e.g., 'newly built structures near a river', 'large vehicle concentrations on open ground'), image-to-image similarity search, automated co-registration, radiometric normalization, false-alarm-suppressed change detection masks, and an interactive GIS exploration dashboard.