ExoVision：AI 驱动的系外行星探测与宜居性评估平台

ExoVision: AI-Driven Exoplanet Detection & Habitability Assessment Platform

Original Author/Maintainer: Kgodiso-Leboho Source: GitHub (https://github.com/Kgodiso-Leboho/ExoVision) Updated: 2026-05-25T21:45:42Z

ExoVision is a full-stack web application integrating astrophysics and machine learning. It automatically identifies potential exoplanets by analyzing light curves, stellar parameters, and planetary features, then assesses their habitability. Key technologies include CNN, LSTM, FastAPI, and React, supporting data from NASA TESS. This platform provides an intelligent tool for exploring data related to extraterrestrial life.

Project Background & Scientific Motivation

Since the first exoplanet around a Sun-like star was discovered in 1995, over 5000 exoplanets have been confirmed. However, detection faces challenges: exoplanets are drowned in stellar light (brightness difference up to 1 billion times).

Traditional detection methods:

• Transit Method: Observe tiny brightness drops when planets pass in front of stars.
• Radial Velocity Method: Detect periodic stellar wobbles due to planetary gravity.

These methods generate massive data, but manual analysis is inefficient. ExoVision aims to automate detection with machine learning and assess habitability of candidates.

Technical Architecture

ExoVision uses a full-stack architecture:

• Frontend: React.js (data upload, visualization with Chart.js/Plotly, result display).
• Backend: FastAPI (data validation via Pydantic, preprocessing pipeline, model inference).
• ML Layer: TensorFlow/PyTorch (multi-model classifier, CNN/LSTM time-series models, habitability assessment algorithm).

Data Models

Planet Transit Features

Feature	Physical Meaning	Scientific Significance
pl_rade	Planet radius (Earth multiples)	Classify planet size (Earth-sized/Super-Earth/Neptune-sized/Jupiter-sized).
pl_orbper	Orbital period (days)	Key parameter for habitable zone position.
pl_insol	Stellar radiation flux (Earth multiples)	Core indicator for surface temperature.
pl_eqt	Equilibrium temperature (Kelvin)	Judge liquid water possibility.

Star Features

Feature	Physical Meaning	Scientific Significance
st_teff	Effective temperature (Kelvin)	Determine star type (M/K/G dwarf).
st_rad	Star radius (Sun multiples)	Calculate planet transit area ratio.

Category Encoding

• Planet size: One-Hot encoded (Earth-sized <1.25R⊕, Super-Earth 1.25-2R⊕, etc.).
• Star temperature: One-Hot encoded (M-dwarf <3500K, K-dwarf 3500-5000K, etc.).

Multi-Model Integration Strategy

ExoVision uses ensemble learning to improve reliability:

1. CNN: Capture local features and periodic patterns in light curves.
2. LSTM: Process time-series data and model long-term dependencies.
3. Traditional ML: XGBoost/Random Forest for structured features.

Model Evaluation Metrics

Metric	Description	Target Value
Accuracy	Overall accuracy	>90%
Precision	Reduce false positives	>85%
Recall	Reduce missed detections	>85%
F1-Score	Harmonic mean of precision/recall	>85%

Precision is critical for reducing false positives (e.g., stellar activity, instrument noise).

Key Habitability Parameters

Based on habitable zone theory:

1. Insolation: Ideal range (0.5-2.0x Earth's). Too low → frozen water; too high → runaway greenhouse effect.
2. Equilibrium Temperature: Ideal range (250-350K, -23°C to 77°C) for liquid water.
3. Planet Size: Earth/Super-Earth are more likely to retain atmosphere/magnetic field.
4. Star Type: K/G dwarfs are optimal (long lifespan, stable activity).

Habitability Scoring

• High: Meets liquid water, suitable temperature, rocky surface.
• Medium: Partial conditions met; need further observation.
• Low: Clearly unsuitable for life.

5-Step User Experience

1. Data Upload: Upload CSV (compatible with NASA Exoplanet Archive/TESS) or use example data.
2. Preprocessing: Auto-handle missing values, outliers, normalization, category encoding.
3. Model Prediction: Classify into candidate exoplanet, false positive, or need further observation.
4. Visualization: Generate light curves, feature importance, prediction confidence, habitability radar charts.
5. Habitability Assessment: Output score, parameter interpretation, comparison with known habitable planets.

Scientific Contributions

1. Accelerate Screening: ML processes thousands of light curves in seconds (100x faster than manual analysis).
2. Reduce False Positives: Learn noise patterns to distinguish real signals.
3. Prioritize Observation: Help astronomers allocate telescope time to high-habitability candidates.

Education & Open Source

• Education: Students can upload real data and learn exoplanet science via visualization.
• Open Source: Provide standardized pipelines, feature engineering methods, and model references.

Current Limitations

1. Data Dependency: Performance depends on training data quality/coverage.
2. Class Imbalance: Real exoplanet signals are rare.
3. Simplified Models: Habitability assessment ignores complex factors (atmosphere, magnetic field).

Future Directions

1. Multi-Task Learning: Predict planet mass, density, atmosphere.
2. Anomaly Detection: Identify non-periodic events (comets, asteroid belts).
3. Transfer Learning: Adapt to PLATO/ARIEL missions.
4. Uncertainty Quantification: Output prediction confidence intervals.
5. Multi-Modal Fusion: Combine radial velocity and direct imaging data.

Conclusion

ExoVision bridges AI and astrophysics, turning massive data into actionable insights. It will play an important role in finding 'another Earth' with next-gen telescopes like JWST. For researchers, students, and enthusiasts, it's a user-friendly tool connecting curiosity to cosmic mysteries.