Machine Learning Land Use Classification from Satelite images

1. Business / Research Question

Problem

Can satellite image patches be automatically classified into land use categories to support environmental monitoring and urban planning?

Motivation

Manual annotation of satellite imagery is costly and time-consuming. Automated classification can:

• Accelerate large-scale analysis
• Support planning and monitoring
• Flag uncertain cases for human review

The system is intended as decision support rather than full automation.

2. Dataset

2.1 Source

We use the EuroSAT RGB dataset:

• ~27,000 satellite image patches
• 10 land use classes
• RGB format, 64×64 resolution
• Images organized in class-based folders

2.2 Classes

• AnnualCrop
• Forest
• HerbaceousVegetation
• Highway
• Industrial
• Pasture
• PermanentCrop
• Residential
• River
• SeaLake

Each image represents a satellite patch labeled with its dominant land use type.

2.3 Dataset Indexing

To prepare the dataset for model training, we construct a master DataFrame containing:

• The image file path
• The corresponding class label

This structured representation will allow us to perform a reproducible train/validation/test split while ensuring consistency across models.

4. Model 1: Random Forest

5. Model 2: CNN

6. Comparison of Results

7. Limitations and Future Improvements