This repository is the home of "SpeckleReducer", a deep learning-based model for speckle noise reduction in SAR satellite imagery, developed as a part of the course 'Advanced Earth Observation' at Wageningen University & Research.
The project explores the application of convolutional neural networks for the task of reducing speckle noise in Sentinel-1 SAR images. The repository provides resources including the preprocessing steps of the SEN12MS dataset, a PyTorch custom dataset class for handling the images, the model architecture built using PyTorch Lightning, as well as scripts used for training, validation, and evaluation. In addition, it contains visualizations and detailed explanations of the results achieved.
This project contributes to the field of remote sensing by proposing a self-supervised approach for speckle noise reduction in SAR imagery, paving the way for future advancements in this area. Please feel free to explore the code and results shown in the paper, and also to contribute to the development of this methodology.