GeoSentinel: Anomaly Detection in Sentinel-2 Time Series

GeoSentinel is a Python-based project designed to detect and visualize anomalies in remote sensing data, specifically from the Sentinel-2 satellite. This tool is valuable for environmental monitoring, change detection, disaster response, and land use tracking.

---

Features

• Download Sentinel-2 time series using the SentinelHub API
• Analyze multi-band data over time for a region of interest (ROI)
• Apply statistical and machine learning methods (e.g., Isolation Forest) for anomaly detection
• Visualize time series and anomaly maps

---

Project Structure

GeoSentinel/
├── sentinel_fetcher.py       # Downloads Sentinel-2 data from SentinelHub
├── anomaly_detection.py      # Detects anomalies in the NDVI or reflectance series
├── visualization.py          # Visualization of time series and anomaly maps
├── config.yaml               # Config file with ROI, time range, bands, credentials
├── main.py                   # Entry point to run the full pipeline
└── README.md                 # Documentation

---

Requirements

Install dependencies with:

pip install -r requirements.txt

Dependencies include:

• numpy, pandas, matplotlib, seaborn
• rasterio, geopandas
• scikit-learn
• sentinelhub

---

Usage

1. Create an account at Sentinel Hub and obtain your credentials.
2. Fill out config.yaml with your AOI, bands, time range, and credentials.
3. Run the pipeline:

python main.py

This will:

• Download and preprocess the time series
• Run anomaly detection
• Output maps and time series plots in the output/ folder

---

Example Output

• NDVI time series with detected outliers
• RGB image overlays with anomaly locations

---

Potential Extensions

• Integrate with deep learning (e.g. LSTM) for sequence-based detection
• Add land cover classification support
• Streamline via Web UI or dashboard

---

License

MIT License For research or environmental monitoring applications. Not affiliated with ESA or Sentinel Hub.