Satellite Land Use and Land Cover Classification

A deep learning-based web application for Satellite Land Use and Land Cover (LULC) classification using a fine-tuned ResNet50 model trained on the EuroSAT RGB dataset.

The project provides a Flask REST API for image classification and a web-based frontend for uploading satellite images and viewing predicted land-cover classes with confidence scores.

Features

• Satellite image classification using ResNet50
• Transfer learning with a pretrained ResNet50 architecture
• 10-class Land Use and Land Cover classification
• Flask REST API for predictions
• Image upload through a web interface
• Confidence scores for predictions
• Probability distribution across all classes
• Docker support for backend deployment
• Health-check API endpoint

Model

The project uses a fine-tuned ResNet50 convolutional neural network.

Model Details

Parameter	Details
Architecture	ResNet50
Learning Approach	Transfer Learning / Fine-Tuning
Input Size	224 × 224 × 3
Output	10-class Softmax
Dataset	EuroSAT RGB
Framework	TensorFlow / Keras
Model Format	.keras

The model performs classification of satellite images into different land-use and land-cover categories.

LULC Classes

The model predicts the following 10 classes:

1. Annual Crop
2. Forest
3. Herbaceous Vegetation
4. Highway
5. Industrial
6. Pasture
7. Permanent Crop
8. Residential
9. River
10. Sea/Lake

Dataset

The model was trained using the EuroSAT RGB dataset, a satellite image dataset containing RGB satellite images covering multiple land-use and land-cover categories.

The dataset contains images from the following classes:

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

System Architecture

Project Structure

satellite-lulc-classification/
│
|--Training/
|    |--LULC_ResNet50_Training  
├── backend/
│   ├── app.py
│   ├── config.py
│   ├── model_loader.py
│   ├── prediction.py
│   ├── requirements.txt
│   ├── Dockerfile
│   ├── .dockerignore
│   └── README.md
│
├── utils/
│   ├── image_utils.py
│   └── response.py
│
├── index.html
│
├── .gitignore
├── .gitattributes
└── README.md

Model Weights

The trained ResNet50 model file is not included in this repository because of its large file size (~350 MB).

This keeps the Git repository lightweight and avoids unnecessarily increasing repository storage.

The backend expects the trained model file:

best_resnet50.keras

Place the model in the location expected by model_loader.py before running the backend.

Note: Model weights are intentionally excluded from GitHub. The repository contains the complete application code required to load and use the model.

Frontend

The project includes an HTML-based frontend for interacting with the classification API.

The frontend:

1. Accepts a satellite image.
2. Sends the image to the Flask API.
3. Receives the model prediction.
4. Displays the predicted land-cover class.
5. Displays the model confidence.
6. Displays probability information.

Technologies Used

• Python
• TensorFlow
• Keras
• ResNet50
• Deep Learning
• Transfer Learning
• Computer Vision
• Flask
• Flask-CORS
• NumPy
• Pillow
• HTML
• JavaScript
• Docker
• Gunicorn

Applications

Satellite LULC classification can support applications such as:

• Land-use monitoring
• Urban development analysis
• Agricultural monitoring
• Forest monitoring
• Environmental analysis
• Remote sensing
• Geographic information systems
• Land management

License

This project is licensed under the MIT License.

Author

Karthik

GitHub: https://github.com/lukarthik67