1D CNN for Sea/Land/Cloud Classification of Hyperspectral Images from the HYPSO-2 Satellite

This project implements the 1D-JustoLiuNet for classification tasks on hyperspectral satellite image captures from the HYPSO-2 satellite. The 1D-JustoLiuNet model is based on recent work by Justo et al. (2023) and optimized for deployment in resource-constrained satellite environments.

Full theoretical background, model design, evaluation metrics, and experimental results are presented in the accompanying final report PDF, submitted as part of an academic research contribution at NTNU SmallSat Lab.

Table of Contents

• Requirements
• Usage
• Structure
• Theory, Testing and Results
• License
• References
• Acknowledgments

Requirements

• CUDA version: 12.6
• UBUNTU version: 22.04
• NVIDIA driver version: 565.57.01

Usage

1. Clone the repository:

   git clone https://github.com/raeesiarya/HYPSO-CNN.git

2. Install required packages:

   pip3 install -r requirements.txt

This project used Python 3.12.3 in a virtual venv environment.

3. Prepare your dataset:

   1. Get the raw data from HYPSO-2:

      python3 scraping/get_data1.py

   2. Run the function create_csv_file in manage_data.py to create CSV files with correct paths.

4. Train and evaluate the model:

   python3 scripts/train.py

Structure

1. Class Diagram (shortened, full version in page 50 here)

Figure 1: Class diagram of the 1D-CNN classification pipeline.

2. Component Diagram

Figure 2: Component of the 1D-CNN classification pipeline.

3. Sequence Diagram

Figure 3: Sequence diagram of the 1D-CNN classification pipeline.

4. Activity Diagram

Figure 4: Activity diagram of the 1D-CNN classification pipeline.

Theory, Testing and Results

All theory, testing and results have been presented in this PDF.

License

This project is licensed under the MIT License. See the LICENSE file for details.

References

• Jon Alvarez Justo, Dennis D. Langer, Simen Berg, Jens Nieke, Radu Tudor Ionescu, Per Gunnar Kjeldsberg, and Tor Arne Johansen.
  Hyperspectral Image Segmentation for Optimal Satellite Operations: In-Orbit Deployment of 1D-CNN.
  Remote Sensing, vol. 17, no. 4, p. 642, 2025. DOI: 10.3390/rs17040642

• Jon A. Justo, Joseph Garrett, Dennis D. Langer, Marie B. Henriksen, Radu T. Ionescu, and Tor A. Johansen.
  An Open Hyperspectral Dataset with Sea-Land-Cloud Ground-Truth from the Hypso-1 Satellite.
  In 2023 13th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS), IEEE, pp. 1–5, 2023.

• Jon Alvarez Justo, Joseph Landon Garrett, Mariana-Iuliana Georgescu, Jesus Gonzalez-Llorente, Radu Tudor Ionescu, and Tor Arne Johansen.
  Sea-Land-Cloud Segmentation in Satellite Hyperspectral Imagery by Deep Learning.
  arXiv preprint arXiv:2310.16210, 2023. arXiv:2310.16210

• Ian Goodfellow, Yoshua Bengio, and Aaron Courville.
  Deep Learning. MIT Press, 2016. Online version

Acknowledgments

This project was developed as part of my work with the NTNU SmallSat Lab. I would like to thank the entire SmallSat team for their collaboration and support throughout the project.