Satellite-Collision-Detection

Advanced satellite collision detection and avoidance system using machine learning, physics-informed models, and standardized orbital data formats (CCSDS CDM, TLE), with explainable AI and optimized maneuver planning.

What's New (Nov 2025)

• Fixed nested folder issue by introducing a flat, well-named structure for explainability code: src/explainability/
• Integrated datasets and formats used in Erik Cupsa's On-Orbit Collision Predictor, including CCSDS CDM sample data
• Added datasets documentation with usage examples and source acknowledgments (datasets/README_DATASETS.md)
• Enhanced About section and project description for clarity

Project Structure

• src/explainability/
  • shap_explainer.py — SHAP-based explainability utilities for model predictions
• datasets/
  • ccsds_cdm_sample.json — sample CCSDS CDM file (Erik Cupsa format)
  • sample_tle_data.csv — sample TLE elements
  • parse_tle.py — helper to parse TLE data
  • README_DATASETS.md — datasets and formats documentation
• models/ — ML and physics-informed models (PINNs, LSTMs, GNNs)
• maneuver_planning/ — delta-V optimization and avoidance planning
• visualization/ — 3D/AR visualization and dashboards

Data Formats and Sources

• CCSDS CDM (Conjunction Data Message)
  • File: datasets/ccsds_cdm_sample.json
  • Fields: TCA, miss distance, relative state, covariance, collision probability, object metadata
  • Source pattern: aligned with Erik Cupsa's repository data format
• TLE (Two-Line Element)
  • Files: datasets/sample_tle_data.csv, parser datasets/parse_tle.py
  • Sources: NORAD/Space-Track, CelesTrak

See datasets/README_DATASETS.md for details, examples, and references.

Explainable AI

This project includes an explainability toolkit to increase operator trust in automated alerts:

• SHAP waterfall and force plots for per-prediction insights
• Global summary plots for feature importance
• Feature ranking by mean |SHAP| value
• Module: src/explainability/shap_explainer.py

Installation

Prerequisites

• Python 3.8+
• Docker (optional for containerized runs)
• Access to telemetry or simulation data (optional)

Setup

# Clone
git clone https://github.com/Anand0295/Satellite-Collision-Detection
cd Satellite-Collision-Detection

# (Optional) Build container
docker build -t satellite-collision-detector .

Usage

Data

• Place additional CCSDS CDM files under datasets/
• Load TLEs with datasets/parse_tle.py

Explainability quick start

import json
import numpy as np
from src.explainability.shap_explainer import SHAPExplainer

# Example model stub with predict(X) -> y_proba
class DummyModel:
    def predict(self, X):
        # Return a probability-like score for demonstration
        return np.clip(0.5 + 0.1 * np.random.randn(len(X)), 0, 1)

model = DummyModel()
background = np.random.randn(100, 10)
explainer = SHAPExplainer(model, background)
X = np.random.randn(5, 10)
feature_names = [f"f{i}" for i in range(X.shape[1])]
ex = explainer.explain_prediction(X, feature_names)
rank = explainer.get_feature_importance_ranking(ex)
print(rank.head())

Project Overview

The project aims to provide accurate, timely conjunction risk assessment and maneuver planning by combining:

• Multi-source data fusion (radar, optical, TLE, CCSDS CDM)
• Advanced ML and physics-informed modeling
• Real-time processing at scale
• Explainable AI for trustworthy decision support
• Operator-focused visualization and APIs for safe maneuver execution

Acknowledgments

• Erik Cupsa — inspiration and data format integration from On-Orbit Collision Predictor
  • https://github.com/Erik-Cupsa/On-Orbit-Collision-Predictor
• NORAD/Space-Track, CelesTrak — orbital element datasets (TLE)
• CCSDS — CDM standard (CCSDS 508.0-B-1)

License

MIT — see LICENSE.