Skip to content
SIH Buddyby Ganeev Singh
Dev
All problem statements
SIH26227Strong pickacceptance 4/5

Semantic Retrieval and Multi-Temporal Change Analysis of Satellite lmagery.

Ministry of defence (MoD) · Space Technology · Software

One of the best-specified problems in the whole set — free public data, a rigorous held-out evaluation, a striking offline demo — but it is won on false-alarm suppression and honest incremental indexing, so build those rather than a semantic search box bolted to a naive change detector.

Data: Public / organiser-generated imagery only (GeoTIFF / COG); no classified data

What it actually is

Satellite archives are huge, and analysts can only search them by metadata — coordinates, date, sensor — which means they must already know where to look. The ask is a system that makes an imagery archive searchable by meaning ('newly built structures near a river') and by change over time, running entirely on-premises and offline, while carefully not reporting seasonal or atmospheric variation as real change.

What to build

A retrieval-and-change platform over a satellite archive with six required capabilities: free-text and image-to-image semantic search over tiles using a remote-sensing foundation model, ranked and filterable by area, date and sensor; multi-temporal change analysis that classifies construction, clearance, water-extent and road development within a time window and estimates the earliest observation supporting each change; false-alarm suppression that treats season, illumination, cloud, haze, shadow and mis-registration as confounders via quality masks and confidence estimates, favouring precision over recall; embedding-based discovery to cluster visually similar sites so one location of interest surfaces others; an analyst review queue with before-and-after evidence, provenance and feedback-driven reranking; and scalable incremental vector indexing that ingests new imagery without a full rebuild — all fully offline with network disabled during evaluation.

Smallest thing that wins the room

With the network physically disabled, type 'newly built structures near a river' and get ranked tiles from the archive, then pick an area and time window and watch the system surface a genuine construction change with before-and-after imagery and a confidence score, while correctly NOT flagging a seasonally greener field as change.

How crowded this one gets

A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.

Moderate90–210 teams expectedroughly 1 in 76–176 wins it

Quieter than 69% of the 240 · #75 of 240 by expected field

A normal-sized field. Your idea has to be good, not miraculous.

Why: defence, intelligence and space bodies drew small fields.

This is a guess, not a fact

Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.

The scores

The number is the shorthand. The line under it is the reason.

What you will be writing

  • Remote-sensing foundation model (Clay / Prithvi / SatCLIP)
  • Vector index (FAISS / Qdrant) incremental ingestion
  • Multi-temporal change detection + classification
  • Confounder suppression (quality masks, radiometric normalisation)
  • GeoTIFF / COG geospatial provenance
  • Offline / air-gapped deployment
  • Satellite image retrieval
  • Change detection
  • Geospatial intelligence

Prior art to read before you start

semantic satellite image search · multi-temporal change analysis · false-alarm suppression in change detection

Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.