SRGAN for Super-Resolution Satellite Images

This project implements Super-Resolution Generative Adversarial Networks (SRGAN) to enhance the resolution of Sentinel-2 satellite imagery using the BigEarthNet-S2 dataset.
The goal is to improve image quality and detail for downstream tasks in remote sensing, land use/land cover analysis, and environmental monitoring.

---

Features

• Preprocessing pipeline for BigEarthNet-S2 (10m, 20m, and 60m spectral bands)
• Custom SRGAN architecture adapted for multi-spectral satellite images
• Training pipeline with GPU acceleration support
• Evaluation using PSNR, SSIM, MSE, and FID metrics
• Visualization tools for comparing low-resolution vs. super-resolved images

---

Dataset

• BigEarthNet-S2: A large-scale benchmark derived from Sentinel-2 images, covering multiple resolutions and spectral bands.