The goal of this repository is to give tools to use Deep Learning on Sentinel-2 satellite images.
A blog post has been written about the context: Fighting Hunger through Open Satellite Data: A New State of the Art for Land Use Classification.
This repository shows:
- How to download Sentinel-2 images using Google Earth Engine Python API.
- How to get a new SOTA (july 2019) on EuroSAT dataset using multispectral images. The model has a 0.99 accuracy (previous SOTA: 0.9857 => 30% less error rate).
- How to use the model pretrained on EuroSAT for you own classification tasks.
- How to process and visualize multispectral images.
You can download the weights of the pretrained model.