You can deploy this model and start earning money today!

KawgKawgKawg/
Manila-Urban-Expansion

Name: Manila-Urban-Expansion
Availability: InStock

Views

No views yet

Manila-Urban-Expansion-Detection

A machine learning web application for predicting urban areas from satellite imagery spectral data. This tool uses a pre-trained Random Forest model to classify urban and non-urban areas based on Landsat spectral features.

🌟 Features 📊 CSV-based Prediction: Upload CSV files with spectral features for urban classification

🎯 Pre-trained Model: Uses a Random Forest classifier trained on Manila urban data

📈 Interactive Visualizations: Multiple charts and graphs for result analysis

📱 Web Interface: User-friendly Gradio interface

📥 Download Results: Export predictions as CSV files

🌍 Spatial Analysis: Optional geographic coordinate support

🎪 Confidence Scoring: Quality assessment for each prediction

🏗️ Technology Stack Technology Purpose Version Python Backend language 3.8+ Gradio Web interface framework ≥3.50.0 Scikit-learn Machine learning library ≥1.0.0 Pandas Data processing ≥1.3.0 NumPy Numerical computations ≥1.21.0 Matplotlib Data visualization ≥3.5.0 Pickle Model serialization Built-in Hugging Face Deployment platform - 📋 Required CSV Format Essential Columns:

B1_coastal,B2_blue,B3_green,B4_red,B5_nir,B6_swir1,B7_swir2,NDVI,NDBI,NDWI,brightness,ratio_swir_nir,ratio_nir_red

Optional Columns:

longitude,latitude # For spatial visualization

Example CSV Structure:

csv

B1_coastal,B2_blue,B3_green,B4_red,B5_nir,B6_swir1,B7_swir2,NDVI,NDBI,NDWI,brightness,ratio_swir_nir,ratio_nir_red
0.123,0.145,0.167,0.189,0.234,0.456,0.378,0.234,0.456,0.123,0.289,1.234,1.456
0.134,0.156,0.178,0.201,0.245,0.467,0.389,0.245,0.467,0.134,0.301,1.245,1.467

🚀 Installation & Setup Local Development: Clone the repository:

bash

git clone <your-repo-url>
cd satellite-urban-prediction

Create virtual environment:

bash

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Install dependencies:

pip install -r requirements.txt

Add your trained model:

bash

# Place your trained model file in the root directory
# File should be named: model.pkl
Run the application:

python final_app.py

🎯 Model Training Information Expected Features: The model expects 13 spectral features in this exact order:

B1_coastal - Coastal aerosol band

B2_blue - Blue band

B3_green - Green band

B4_red - Red band

B5_nir - Near Infrared band

B6_swir1 - Short-wave Infrared 1

B7_swir2 - Short-wave Infrared 2

NDVI - Normalized Difference Vegetation Index

NDBI - Normalized Difference Built-up Index

NDWI - Normalized Difference Water Index

brightness - Average brightness

ratio_swir_nir - SWIR to NIR ratio

ratio_nir_red - NIR to Red ratio

Model Architecture: Algorithm: Random Forest Classifier

Trees: 100 estimators

Max Depth: 10 levels

Training Data: Manila urban/rural areas

Accuracy: >85% on test data

📊 Output Results Visual Outputs: Prediction Distribution - Bar chart of urban vs non-urban predictions

Probability Distribution - Histogram of prediction confidence

Spatial Distribution - Geographic plot (if coordinates provided)

Confidence Levels - Quality assessment of predictions

Data Outputs: Prediction Label: Urban/Non-Urban classification

Probability Score: Confidence score (0-1)

Confidence Level: Qualitative assessment (Low/Medium/High/Very High)

Geographic Coordinates: If provided in input

Model Results:

We evaluated five models (Logistic Regression, Random Forest, Decision Tree, K-Nearest Neighbors, and SVC) using Randomized Search with 5-fold cross-validation, optimizing for F1 score.
Best Model: Random Forest Classifier
Best Parameters: 200 estimators, max depth = 20, min samples split = 5, min samples leaf = 4
Performance: Accuracy (0.9996), Precision (0.9965), Recall (1.0000), F1-Score (0.9982), ROC-AUC (1.0000)
The Random Forest model demonstrated near-perfect performance with only 4 errors out of 10,000 samples (0.04%), all of which were false urban classifications. No false non-urban errors were observed.

This indicates that the model is highly reliable for detecting urban expansion in Manila, though a slight threshold adjustment may reduce false positives further.

Downloadable Files: Complete results CSV with all predictions

Preserves all original input data plus predictions

🎮 How to Use Prepare Your Data:

Collect spectral data from Landsat imagery

Calculate required indices (NDVI, NDBI, NDWI)

Format as CSV with expected column names

Run Prediction:

Upload CSV file through the web interface

Click "Predict Urban Areas"

View interactive results and visualizations

Analyze Results:

Review prediction statistics

Examine confidence levels

Download results for further analysis

Interpret Results:

Urban areas: High NDBI, moderate brightness

Non-urban: High NDVI (vegetation) or other features

Confidence scores indicate prediction reliability

🔧 Customization Modifying Expected Features: Edit the expected_features list in app.py:

python

expected_features = [
    'B1_coastal', 'B2_blue', 'B3_green', 'B4_red', 
    'B5_nir', 'B6_swir1', 'B7_swir2',
    'NDVI', 'NDBI', 'NDWI', 'brightness', 
    'ratio_swir_nir', 'ratio_nir_red'
]

Adding New Visualizations: Extend the plotting section in predict_urbanization_csv() function:

python

# Add new subplot
ax5 = plt.subplot(2, 3, 5)  # Adjust grid as needed
ax5.plot(new_data)
ax5.set_title('New Visualization')

Model Replacement: Replace model.pkl with your new model file. Ensure it has:

.model attribute: Trained classifier

.scaler attribute: Fitted StandardScaler

.feature_names attribute: List of expected features

🐛 Troubleshooting Common Issues: Missing Model File:

text

❌ Pickle file urban_model.pkl not found
Solution: Ensure urban_model.pkl is in the root directory

CSV Format Error:

text

❌ Missing features in CSV: B5_nir, NDVI, ...
Solution: Check column names match expected features

text

Memory Issues:
Solution: Reduce sample size or upgrade Hugging Face Space hardware

text

Visualization Errors:
Solution: Check for NaN values in input data

text

Performance Tips:
Use smaller CSV files for testing (<10,000 rows)

Pre-calculate spectral indices before upload

Ensure numeric columns don't contain text values

Handle missing values before upload

📈 Example Use Cases Urban Planning: Monitor urban expansion over time

Identify potential development areas

Assess urban density patterns

Environmental Research:

Study urban heat island effects
Analyze vegetation loss in urban areas
Monitor water body changes near cities

Academic Projects:

Remote sensing coursework
Machine learning demonstrations
Geographic information systems (GIS) studies

🤝 Contributing Fork the repository

Create a feature branch

Make your changes

Test thoroughly

Submit a pull request

Development Priorities: Add support for multiple model types

Implement batch processing for large files

Add temporal analysis capabilities

Include more visualization options

Support for additional satellite data formats

📝 License This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments Landsat Program for satellite imagery data

Scikit-learn team for machine learning tools

Gradio team for the web framework

Hugging Face for deployment platform

📞 Support For questions and support:

Check the troubleshooting section above

Review example CSV formats

Ensure model file is properly formatted

Verify all dependencies are installed

⭐ If you find this project useful, please give it a star on GitHub!

Built with ❤️ for urban planning, environmental research and financial forecasting

KawgKawgKawg/Manila-Urban-Expansion

Manila-Urban-Expansion-Detection

KawgKawgKawg/
Manila-Urban-Expansion