A Real-World Benchmark for Sentinel-2 Multi-Image Super-Resolution.

Area of Science:

• Remote Sensing
• Computer Vision
• Image Processing

Background:

• Limited spatial resolution of satellite imagery (e.g., Sentinel-2 at 10m) restricts applications, especially when higher resolution is costly or infeasible.
• Multi-image super-resolution leverages information fusion from multiple satellite revisits to enhance image reconstruction accuracy.
• Existing benchmarks often rely on simulated data, which may not accurately represent real-world operating conditions for super-resolution tasks.

Purpose of the Study:

• To introduce MuS2, a novel benchmark dataset for super-resolving multiple Sentinel-2 images.
• To provide the first end-to-end evaluation procedure for multi-image super-resolution using Sentinel-2 data.
• To facilitate advancements in the state-of-the-art for super-resolution algorithms in remote sensing.

Main Methods:

• Development of the MuS2 benchmark dataset using WorldView-2 imagery as high-resolution reference.
• Establishment of an end-to-end evaluation framework for multi-image super-resolution.
• Utilizing super-resolution algorithms for enhancing Sentinel-2 images through information fusion.

Main Results:

• The MuS2 benchmark offers a realistic dataset for evaluating super-resolution algorithms on Sentinel-2 imagery.
• The proposed evaluation procedure enables comprehensive assessment of algorithm performance.
• The benchmark is expected to drive progress in multi-image super-resolution techniques.

Conclusions:

• The MuS2 benchmark addresses the scarcity of real-world data for multi-image super-resolution.
• The established evaluation procedure will standardize and accelerate research in this domain.
• This work is anticipated to significantly contribute to improving the spatial resolution of satellite imagery.