Showcase your innovative solution for in-orbit data processing and run it onboard the Φsat-2 mission.

M1 - Team profile

M2 - Idea concept

M3 - Initial model

Final submission (Pitch + Final Prototype)

OrbitalAI Φsat-2

# Starter toolkit

## Download our ready-to-use dataset 

To help you get started with the challenge, we provide a ready-to-use dataset of 490 simulated Φsat-2 images with locations across the globe. Instructions on how to directly download the pre-computed core dataset are available in this [Jupyter notebook](https://github.com/AI4EO/orbitalAI/tree/main/phisat-2) ([phisat-2-simulator.ipynb](https://github.com/AI4EO/orbitalAI/blob/main/phisat-2/phisat-2-simulator.ipynb)) on GitHub. Information about the dataset contents is available in the [README_Phisat-2.md](https://s3.waw2-1.cloudferro.com/swift/v1/AUTH_afccea586afd4ef3bb11fe37dd1ddfbf/OrbitalAI_Data/README_Phisat-2.md) file. 

An example patch of simulated Phisat-2 image patches can be directly downloaded [here](https://s3.waw2-1.cloudferro.com/swift/v1/AUTH_afccea586afd4ef3bb11fe37dd1ddfbf/OrbitalAI_Data/335040-5955400_32618_2019-10-12T16-28-28.zip). 

The dataset is available for direct download. Consult the last section (Core dataset) of the "phisat-2-simulator.ipynb" [Jupyter notebook](https://github.com/AI4EO/orbitalAI/tree/main/phisat-2) on how to download them. 

## Simulate your own image data 

To help you get started with the challenge, we prepared a [Jupyter notebook](https://github.com/AI4EO/orbitalAI) ([phisat-2-simulator.ipynb](https://github.com/AI4EO/orbitalAI/blob/main/phisat-2/phisat-2-simulator.ipynb)) with instructions and code to simulate the Φsat-2 satellite images that you can train your models on. The image simulation processing steps are outlined in the [Phisat2_Mission_Overview](https://ai4eo.eu/wp-content/uploads/2023/02/Phisat-2_Mission_Overview_Web.pdf). The notebook contains the workflow to simulate the Φsat-2 images starting from Sentinel-2 imagery and using [eo-learn](https://eo-learn.readthedocs.io/en/latest/index.html), a Python library specifically designed for Earth Observation data.  The output imagery is saved in the GeoTiff file format. 

## How to download Sentinel-2 images 

The entire archive of Sentinel-2 imagery is available through Sentinel Hub. To access it, you can create a [free trial account](https://apps.sentinel-hub.com/dashboard/), please find more information on the configuration of Sentinel Hub [here](https://sentinelhub-py.readthedocs.io/en/latest/configure.html).  

## Available computing resources 

A limited amount of machine learning computing resources will be made available to challenge participants. The resources (CPU/GPU) are kindly provided by the ESA Network of Resources (NoR) and can be accessed at: https://orbitalai.hub.eox.at/.

JupyterLab provides an Interactive Development Environment (IDE) where users can develop their own applications in the Python language using Jupyter Notebooks. In addition, MLflow integration provides a graphical interface for tracking experiments and parameters for machine learning projects. MLflow offers a simple implementation for the majority of ML libraries and models.

By default, you have one profile using 8GB 2CPUs offering standard CPU capabilities. The ready-to-use dataset is also available there. On request, you can get additionally a GPU profile, the best option for developing large models.

More information about the computing resources available and limitations can be found on the platform upon registration and validation.  

Access to these resources will be provided pending the validation by the organizing team, at its sole and absolute discretion, favouring European teams, start-ups, universities, institutions, SMEs, and participants from Global South.

OrbitalAI Φsat-2

Contact Us

Follow Us