MalaGIS

Sharing GIS Technologies, Resources and News.

Remote sensing imagery is crucial for global monitoring but often limited by sensor spatial resolution and the high cost of acquiring ultra-high-resolution data. The Sentinel-2 (S2) mission provides multispectral imagery across 13 bands at 10m, 20m, and 60m resolutions. However, these resolutions may not capture fine details required for tasks like land cover mapping, agricultural monitoring, or disaster assessment. Super-Resolution (SR) technology addresses this by reconstructing high-resolution images from low-resolution inputs, significantly enhancing spatial detail in S2 imagery for more precise data support.

SR4RS (Super-Resolution for Remote Sensing)

Key Features:

• Open-source software based on Orfeo ToolBox TensorFlow (OTBTF) module
• Utilizes CNN models (e.g., ESRGAN)
• Pretrained model based on 250 paired Spot-6/7 and S2 images from 2020
• Enhances 10m resolution bands (B2/B3/B4/B8) to 2.5m

Operation Guide:

1. Pull Docker image

   docker pull mdl4eo/otbtf:latest

2. Run container (mount data directory)

   docker run --rm -ti -v /yourlocalpath:/data mdl4eo/otbtf:latest

3. Download pretrained model

wget https://nextcloud.inrae.fr/s/boabW9yCjdpLPGX/download/sr4rs_sentinel2_bands4328_france2020_savedmodel.zip
unzip sr4rs_sentinel2_bands4328_france2020_savedmodel.zip

4. Band stacking (B4/B3/B2/B8 sequence)

   otbcli_ConcatenateImages -il band4.tif band3.tif band2.tif band8.tif -out image_4328.tif

5. Execute super-resolution processing

   python sr4rs/code/sr.py \
     --savedmodel sr4rs_sentinel2_bands4328_france2020_savedmodel \
     --input /yourpath/S2_image/images_4328.tif \
     --output sr_output.tif

   Output​​: Generates 2.5m resolution imagery

S2DR3 (Sentinel-2 Deep Resolution 3.0)

​​
Breakthrough Advantages​​:

• Enhances all 12 spectral bands to 1m resolution
• Custom ANN architecture preserves soil/vegetation spectral characteristics
• Reconstructs objects with minimum 3m spatial features
• Outputs include: True-color RGB, NDVI, 10-band multispectral, infrared pseudo-color imagery
  ​​
  Workflow​​:

1. Open Google Colab notebook
2. Input target area center coordinates
3. Automatically generates processed results

S2DR3 demonstrates superior geometric object representation, organizing building clusters of varying sizes into clearly defined rectangular structures.

Technical Comparison

Conclusion

S2DR3 leads in resolution and multispectral processing, particularly excelling in geometric regularization of urban structures. SR4RS offers greater flexibility for research customization through its open-source nature and trainable architecture. Together, these technologies advance the application of Sentinel-2 imagery in precision agriculture and disaster response scenarios.

Reference

https://www.geodose.com/2025/07/sentinel-2-super-resolution-sr4rs-s2dr3.html