SIH26142: Deep Learning Based Super Resolution Mapping (SRM) from Medium Resolution Satell — SIH 2026 | Wynit

SIH26142: Deep Learning Based Super Resolution Mapping (SRM) from Medium Resolution Satell — SIH 2026 | Wynit

What they'll ask you

11

Generated from the brief's own wording and the competition's published rules, never from a guess about what this ministry prefers.

Specific hardware requirements for training and inference?
The brief never answers this, so a panel will. Whatever you decide, say it the same way twice.
Exact evaluation metrics for accuracy assessment?
The brief never answers this, so a panel will. Whatever you decide, say it the same way twice.
Has any part of this been shown at a previous event, hackathon or college project?
The guidelines are explicit: your solution must not have appeared in any previous event or programme, of any sort. A recycled project is what a team under time pressure reaches for.

8 morethe ones a panel gets to if there is time

The brief asks for image pre-processing module. How would you build that?
Decoded from SIH26142 itself: Takes medium-resolution Sentinel-2 satellite imagery and prepares it for the generative model. The brief asks for it by name.
The brief asks for deep learning super-resolution model. How would you build that?
Decoded from SIH26142 itself: Applies a trained generative model like a GAN, diffusion model, or transformer to enhance spatial resolution to under 4 meters. The brief asks for it by name.
The brief asks for uncertainty and validation module. How would you build that?
Decoded from SIH26142 itself: Validates enhanced outputs against high-resolution reference data and accounts for model-inferred error components. The brief asks for it by name.
The brief asks for application support output. How would you build that?
Decoded from SIH26142 itself: Produces final enhanced images suitable for crop monitoring, urban analysis, and disaster assessment. The brief asks for it by name.
Why not use what already exists? Name the closest thing to this that is already running.
A team that has not named the alternative themselves is answering this for the first time in the room.
Which single thing will you demonstrate end to end, start to finish, with nothing skipped?
Ours, not a rule: a narrow thing that fully works survives questioning better than a broad thing that half works. If nobody on the team can name it, that is the finding.
Show me this working: the jury will check whether your model transforms 10m Sentinel-2 imagery into products under 4m resolution while preserving geospatial and spectral consistency.
This is the evaluator read for your problem statement, decoded from the brief's own wording.
Show me this working: they will also test how well you handle model uncertainty and validate your outputs against high-resolution reference data.
This is the evaluator read for your problem statement, decoded from the brief's own wording.

The government's original description (435 words)

• Background Medium-resolution satellite imagery, typically ranging from 10 to 30 meters, is widely used in change detection, agriculture, land-cover mapping, disaster monitoring, and urban planning because it offers broad coverage and frequent revisit time. However, the spatial detail is often insufficient for fine-scale analysis, such as identifying small buildings, narrow roads, field boundaries, or localized damage assessment. This creates a need for advanced deep learning based generative enhancement techniques that can extract greater value from existing Earth observation data. • Description Medium-resolution satellite imagery, usually ranging from 10 to 30 meters, is widely used in remote sensing for agriculture monitoring, land-cover mapping, urban planning, disaster assessment, and environmental observation because it provides large-area coverage and frequent revisit capability. However, its spatial resolution is often not sufficient to clearly identify fine details such as narrow roads, small buildings, field boundaries, water edges, or localized damage. This limitation reduces the accuracy and confidence of interpretation and decision-making in applications that require detailed ground-level information. Generative AI super-resolution addresses this problem by using advanced models such as GANs, diffusion models, and deep neural networks to enhance medium-resolution satellite images into sharper and more information-rich finer outputs. These models learn spatial textures, patterns, edges, and spectral relationships from training data containing both medium-resolution and high-resolution image pairs. The goal is not simply to make the image visually clearer, but to reconstruct useful fine-scale details while preserving the original geographic and spectral consistency of the satellite data. The expected solution is a robust AI-based super-resolution framework that can take medium-resolution satellite imagery as input, perform pre-processing, apply a trained generative model, and produce an enhanced spatial resolution image, suitable for analysis. The system should improve feature visibility, support better classification, change detection, crop monitoring, urban mapping, and disaster response. At the same time, it must clearly manage uncertainty because some reconstructed details are inferred by the model and not directly observed. Therefore, validation against high-resolution reference data is essential to ensure that the enhanced outputs are scientifically reliable and useful for real-world remote sensing applications. • Expected Solution The expected solution is a robust super-resolution framework model based on the choice of participating team (Transformers/Generative/CNN etc.) that can transform the input medium-resolution satellite imagery (10m Sentinel-2 Satellite Imagery) into sharper, information-rich products (<4m) while preserving geospatial and spectral consistency. The solution should include pre-processing, model training with paired datasets, accuracy assessment, and validation against high-resolution references. Ideally, it should support applications such as crop monitoring, urban analysis, and disaster assessment. The final outcome should improve in-terms of interpretability and analytical utility, while clearly accounting for uncertainty and error components.