Skip to content
SIH Buddyby Ganeev Singh
Dev
All problem statements
SIH26142Worth consideringacceptance 3/5

Deep Learning Based Super Resolution Mapping (SRM) from Medium Resolution Satellite Imageries

National Technical Research Organisation (NTRO) · Smart Education · Software

The demo is striking and the data is free, but this problem is won on proving your output reconstructs real detail rather than hallucinating it — build the geographic-fidelity evaluation, because that is exactly where an NTRO judge will probe.

Open dataset ↗

What it actually is

Free satellite imagery at ten to thirty metres covers wide areas frequently but is too coarse to make out small buildings, narrow roads or field boundaries. The ask is a deep learning super-resolution model that sharpens medium-resolution satellite images into finer detail — not just prettier, but reconstructing genuine fine-scale features while keeping the geographic and spectral data honest.

What to build

A super-resolution framework taking medium-resolution satellite imagery, such as Sentinel-2 at ten metres, and reconstructing finer spatial detail using a GAN, diffusion model or deep network trained on paired medium and high-resolution imagery, with the critical constraint that outputs preserve geographic registration and spectral consistency rather than hallucinating plausible but false detail, evaluated both on standard image-quality metrics and on whether downstream tasks like building or road extraction genuinely improve on the enhanced imagery.

Smallest thing that wins the room

Take a real Sentinel-2 tile, super-resolve it, and show alongside the true high-resolution image of the same location that your output recovers the actual road and field-boundary structure rather than inventing convincing but wrong detail, with the quantitative fidelity metrics reported.

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.

Moderate150–340 teams expectedroughly 1 in 123–285 wins it

Quieter than 39% of the 226 · #138 of 226 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

  • Real-ESRGAN / SwinIR / diffusion super-resolution
  • Sentinel-2 imagery via Copernicus Data Space
  • Paired data construction with co-registration
  • Spectral consistency loss terms
  • Downstream task evaluation (building/road extraction)
  • SSIM / LPIPS + geographic fidelity metrics
  • Satellite super-resolution
  • Generative deep learning
  • Remote sensing

Prior art to read before you start

satellite image super-resolution · generative detail reconstruction with fidelity constraints · medium-to-high resolution enhancement

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.