CubeSat_ImageClassify

Description

Welcome to the CubeSat Image Classify Hackathon challenge! This challenge includes the following Notebooks:

• Notebook 1: Introduction to the problem and an overview of the hackathon
• Notebook 2: Reading and analyzing the astronomical data
• Notebook 3: Classification using a machine learning model
• Notebook 4: Classification using a deep learning model
• Notebook 5: Evaluation

Data

The data used in this hackathon can be found at link . It contains approximately 16,000 images, each with a size of 3x512x512. The images are classified into the following categories:

0. Blurry
1. Corrupt
2. Missing_Data
3. Noisy
4. Priority

Hackathon Task

Develop a machine learning model that accurately classifies data captured by CubeSats. The goal is to prioritize which images are most valuable for transmission back to Earth, given the limited onboard resources and slow data downlink speeds. Your task is to create a lightweight model that improves the efficiency and/or classification accuracy of the existing solution in this paper.

Video

• In addition to reviewing the provided notebooks, we recommend watching this video link. In the video, Prof. Thron discusses the challenge in detail and explores possible approaches to achieving an optimal solution.

• We also recommend watching this video, if you plan to use ilifu, our cloud computing system, during the hackathons.

Prerequisites

All the necessary libraries and dependencies to run the notebooks are listed in the requirements.txt file.

Installation

On Your Local Machine

To install a single package, use the following command:

pip install --user <package>

To install all required packages, run:

pip install -r requirements.txt

Would you like to clone this repository? Feel free!

git clone https://github.com/Hack4Dev/CubeSat_ImageClassify.git

Then make sure you have the right Python libraries for the notebooks.

New to Github?

The easiest way to get all of the lecture and tutorial material is to clone this repository. To do this you need git installed on your laptop. If you're working on Linux you can install git using apt-get (you might need to use sudo):

apt install git

You can then clone the repository by typing:

git clone https://github.com/Hack4Dev/CubeSat_ImageClassify.git

To update your clone if changes are made, use:

cd CubeSat_ImageClassify/
git pull

Original research work:

Keenan A. A. Chatar, Ezra Fielding, Kei Sano, and Kentaro Kitamura "Data downlink prioritization using image classification on-board a 6U CubeSat", Proc. SPIE 12729, Sensors, Systems, and Next-Generation Satellites XXVII, 127290K (19 October 2023); https://doi.org/10.1117/12.2684047.

Data used

• For the original dataset, please visit link
• If you prefer to access the tutorial more quickly, you can use this link; however, please note that the download time will be longer.