Skip to content

README.md

Latest commit

 

History

History
203 lines (139 loc) · 5.53 KB

README.md

File metadata and controls

203 lines (139 loc) · 5.53 KB
Raw

Drone Detection System

This system uses a Software Defined Radio (SDR) with machine learning to detect drones in the vicinity.

System Requirements

  • Python 3.6+
  • PlutoSDR or compatible SDR hardware
  • libiio library

Installation

  1. Clone this repository:
git clone https://github.com/yourusername/drone-detection-system.git
cd drone-detection-system
  1. Install required Python packages:
pip install -r requirements.txt
  1. Connect your PlutoSDR device via USB or network.

Using the System

1. Collecting Training Data

To train an accurate model, you'll need to collect both drone data and background/noise data:

# To collect drone data (with drone flying nearby):
python drone_detection_system.py --collect-data --num-samples 1000

# To collect background data (with no drone):
python drone_detection_system.py --collect-data --num-samples 1000

After collecting background data, rename the output file:

mv drone_dataset/drone_data_*.npz drone_dataset/background_*.npz

2. Training the Model

You have two options for training models:

Basic Training

python train_drone_model.py

Advanced Training with Multiple Models

python advanced_model_training.py

The advanced training script:

  • Uses more sophisticated features
  • Trains multiple model types (XGBoost, Random Forest, SVM)
  • Performs hyperparameter tuning
  • Handles class imbalance with SMOTE
  • Generates performance comparisons
  • Selects the best model automatically

3. Running the Detection System

Start the detection system:

python drone_detection_system.py

This will:

  • Connect to your SDR
  • Load the trained model
  • Display a real-time waterfall display
  • Show drone detection probabilities
  • Alert when a drone is detected

Command-line Options

Data Collection

python drone_detection_system.py --collect-data [--num-samples N]
  • --collect-data: Enables data collection mode
  • --num-samples N: Number of samples to collect (default: 1000)

Detection System

python drone_detection_system.py [--tx-channel N] [--rx-channel N]
  • --tx-channel N: Transmit channel to use (default: 0)
  • --rx-channel N: Receive channel to use (default: 0)

Basic Model Training

python train_drone_model.py [--dataset-dir DIR] [--output FILE] [--test-size S]
  • --dataset-dir DIR: Directory containing the dataset (default: 'drone_dataset')
  • --output FILE: Output model filename (default: 'xgboost_drone_detection_model.json')
  • --test-size S: Fraction of data to use for testing (default: 0.2)

Advanced Model Training

python advanced_model_training.py [--dataset-dir DIR] [--output-dir DIR] [--test-size S] [--no-smote] [--no-advanced-features]
  • --dataset-dir DIR: Directory containing the dataset (default: 'drone_dataset')
  • --output-dir DIR: Directory to save models and plots (default: 'models')
  • --test-size S: Fraction of data to use for testing (default: 0.2)
  • --no-smote: Disable SMOTE class balancing
  • --no-advanced-features: Use simple feature extraction instead of advanced features

Feature Engineering

The system extracts advanced features from the radio signals:

  1. Frequency Domain Features:

    • Spectral statistics (mean, std, skewness, kurtosis)
    • Spectral entropy
    • Peak analysis

  2. Time Domain Features:

    • Statistical features of I/Q components
    • Zero-crossing rate
    • Envelope analysis

  3. Modulation Features:

    • AM component detection
    • FM/PM component detection

  4. Drone-Specific Features:

    • Control band energy ratios
    • Frequency hopping detection

  5. Wavelet Features:

    • Multi-resolution time-frequency analysis

Tips for Better Detection

  1. Collect diverse data: Include drone data from different drones at various distances and angles.

  2. Include environmental variations: Collect background data in different environments and conditions.

  3. Use advanced training: The advanced training script will typically yield better results than the basic one.

  4. Try different models: The system will compare XGBoost, Random Forest, and SVM models to find the best one.

  5. Adjust SDR parameters: Try different gain settings and sample rates for optimal performance.

  6. Hardware placement: Place your SDR antenna with clear line of sight for best reception.

Troubleshooting

SDR Connection Issues

If you have trouble connecting to your SDR:

  1. Check USB connection or network settings
  2. Ensure the SDR is powered properly
  3. Verify libiio is installed correctly
  4. Try unplugging and reconnecting the device

Performance Issues

If the system is running slowly:

  1. Reduce the sample rate
  2. Close other applications consuming CPU resources
  3. Adjust buffer sizes for better performance
  4. Use simpler feature extraction with --no-advanced-features

Detection Accuracy

If detection accuracy is poor:

  1. Collect more training data from your specific drone
  2. Ensure training data includes both drone and background samples
  3. Adjust detection threshold in the code (PROBABILITY_THRESHOLD in drone_detection_system.py)
  4. Try running the advanced model training

Training Failures

If model training fails:

  1. Make sure you have sufficient drone and background samples
  2. Check that PyWavelets is installed if using advanced features
  3. Try using simpler features with --no-advanced-features
  4. Disable SMOTE with --no-smote if you encounter memory errors

License

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