Machine-learning experiments on EuroSAT satellite imagery and Salinas hyperspectral data, covering preprocessing, supervised classification and evaluation.
The problem
Remote-sensing inputs combine spatial structure with high-dimensional spectral information, so data preparation and evaluation matter as much as the choice of model.
What was built
Notebook-based machine-learning experiments on EuroSAT satellite imagery and the Salinas hyperspectral dataset, with their labels.
Technical approach
- Load remote-sensing imagery and labels
- Prepare high-dimensional image features
- Run supervised classification experiments
- Evaluate and compare the output of the experiments
- Visualise scientific and geospatial results where they were generated
Engineering decisions
- Keep this work framed as scientific ML experimentation, not a deployed product.
- Separate verified dataset facts from model details that still need a notebook review.
- Hold back exact metrics until the full evaluation output has been checked.
How it was checked
- The repository confirms EuroSAT imagery.
- The repository confirms Salinas hyperspectral data and labels.
- Machine-learning notebooks and experiments are present.
What it shows
Machine learning applied to scientific and geospatial imagery, not only small tabular examples.
About the source
Academic and R&D project. No production or commercial remote-sensing claim is made.