🛰️ Space Debris AI — Intelligent Orbital Debris Detection & Visualization

An AI-powered system that detects space debris from satellite images using YOLOv8 and visualizes debris orbiting Earth through an interactive 3D dashboard.

Live Demo: [Streamlit Cloud Deployment Link]

---

✨ Features

• 🤖 AI Debris Detection — YOLOv8-based object detection for space debris in satellite imagery
• 🌍 Interactive 3D Earth — Three.js globe with animated orbital debris, color-coded by risk level
• 📊 Real-time Analytics — Debris count, confidence scores, risk analysis, altitude/velocity charts
• 🧠 Model Performance Dashboard — mAP, precision, recall, F1 score, confusion matrix, P/R curves
• 📋 Detection History — Persistent log of all detection sessions with trend analysis
• 🔬 Synthetic Dataset Generator — Procedural generation of labeled satellite imagery for training
• ✅ Unit Tests — 55 tests covering inference, preprocessing, and orbital simulation modules

🧠 Deep Learning Pipeline

Synthetic Dataset Generator → YOLOv8n Training (30 epochs) → Model Evaluation → Dashboard
       ↓                              ↓                            ↓
  300 labeled images          debris_detector.pt           mAP, P/R curves,
  (YOLO format)               (trained weights)           confusion matrix

Architecture

• Model: YOLOv8 Nano (single-stage real-time object detector)
• Optimizer: AdamW with learning rate 0.001
• Input: 640×640 satellite images
• Output: Bounding boxes + confidence scores for debris objects
• Augmentation: Random rotation, brightness, contrast, noise, horizontal flip

Performance Metrics (30 Epochs)

• mAP@50: 0.9538
• mAP@50-95: 0.6391
• Precision: 0.9325
• Recall: 0.9044
• F1 Score: 0.9182

The model generalizes exceptionally well on our synthetic debris dataset, maintaining over 90% precision and recall for high-confidence object identification.

🚀 Quick Start

# 1. Install dependencies
pip install -r requirements.txt

# 2. Generate training dataset
python dataset/generate_dataset.py

# 3. Train the model
python training/train.py --epochs 30

# 4. Launch the dashboard
streamlit run dashboard/app.py

Open http://localhost:8501 in your browser.

📁 Project Structure

space-debris-ai/
├── dashboard/              # Streamlit web application
│   └── app.py              # Main app (5 tabs: Detection, 3D, Analytics, History, Model)
├── dataset/                # Training dataset (YOLO format)
│   ├── dataset.yaml        # YOLO dataset config
│   ├── generate_dataset.py # Synthetic data generator
│   ├── train/              # Training images + labels
│   ├── valid/              # Validation images + labels
│   └── test/               # Test images + labels
├── inference/              # Model inference pipeline
│   └── predict.py          # Detection with YOLO or demo fallback
├── models/                 # Trained model weights
│   └── debris_detector.pt  # Trained YOLOv8 weights
├── training/               # Model training scripts
│   └── train.py            # Training + evaluation + metrics export
├── visualization/          # 3D Earth & orbital simulation
│   ├── earth_3d.py         # Three.js WebGL globe generator
│   └── orbital_simulation.py
├── utils/                  # Helper utilities
│   ├── preprocessing.py    # Data augmentation pipeline
│   └── visualization.py    # Detection drawing utilities
├── tests/                  # Unit tests (55 tests)
│   ├── test_predict.py
│   ├── test_preprocessing.py
│   └── test_orbital_simulation.py
├── logs/                   # Training metrics & detection logs
│   └── training_metrics.json
├── requirements.txt
└── README.md

🏋️ Training Your Own Model

# Generate synthetic dataset (customizable count)
python dataset/generate_dataset.py --train 300 --valid 75 --test 75

# Train with custom parameters
python training/train.py --epochs 50 --batch 16 --imgsz 640 --lr 0.001

# Model is saved as models/debris_detector.pt
# Metrics are saved to logs/training_metrics.json

🧪 Running Tests

pip install pytest
python -m pytest tests/ -v

🛠️ Tech Stack

Component	Technology
Object Detection	YOLOv8 (Ultralytics) + PyTorch
3D Visualization	Three.js (WebGL)
Dashboard	Streamlit
Charts	Plotly
Image Processing	OpenCV, Pillow
Data Handling	Pandas, NumPy
Testing	pytest

📄 License

MIT