Skip to content

Latest commit

 

History

History
 
 

meteor_detect_class

README.md

Meteor Burst Detection and Classification Project

This project comprises two main Python scripts that work together to detect and classify meteor bursts from audio spectrograms.

Please see also documentation.

Overview

  1. prime_detection.py

    • Captures audio in 30-second segments using the TwitchRealtimeHandler.
    • Creates spectrograms with the plot_spectrogram function.
    • Processes the spectrograms to detect and classify meteor bursts.

  2. detector_and_classification.py

    • Contains the detect_and_cluster_bursts function.
    • Uses ORB feature detection and DBSCAN clustering to identify meteor bursts.
    • Classifies bursts into:
      • Critical bursts: Bursts lasting at least 0.5 seconds.
      • Non-critical bursts: Bursts lasting less than 0.5 seconds.

prime_detection.py

Key Features

  • Audio Recording: Captures live audio streams from a Twitch source.
  • Spectrogram Generation: Converts audio data into spectrogram images for processing.
  • CSV Logging: Logs results, including the count of critical and non-critical bursts, to a daily CSV file.

Workflow

  1. Audio Capture:

    • Records 30-second audio segments from a specified Twitch stream.
    • Saves the audio as a WAV file.

  2. Spectrogram Generation:

    • Plots and saves a spectrogram image from the recorded audio.
    • Dynamically adjusts the noise threshold based on spectrogram analysis.

  3. Burst Detection:

    • Calls detect_and_cluster_bursts from detector_and_classification.py.
    • Updates hourly and daily burst counts.

  4. Daily CSV Logging:

    • Records timestamped results to a CSV file.
    • Creates a new CSV file each day.

detector_and_classification.py

Key Features

  • ORB Detection:

    • Uses the ORB algorithm to detect keypoints in the spectrogram image.

  • DBSCAN Clustering:

    • Groups keypoints into clusters.
    • Labels clusters as bursts.

  • Burst Classification:

    • Measures the duration of each burst.
    • Classifies bursts based on a duration threshold of 0.5 seconds.

Workflow

  1. Feature Detection:

    • Reads a spectrogram image.
    • Detects keypoints using the ORB algorithm.

  2. Clustering:

    • Applies DBSCAN clustering to group keypoints.
    • Visualizes clusters with bounding boxes (green for critical, red for non-critical).

  3. Classification:

    • Classifies bursts based on their duration.
    • Returns lists of critical and non-critical bursts.

Output

  • Annotated spectrogram image with detected bursts.
  • Lists of critical and non-critical bursts.

Installation and Run

Prerequisites

  • Python 3.8+
  • Required libraries: numpy, opencv-python, matplotlib, scikit-learn, pyaudio, wave, twitchrealtimehandler, pandas.

Setup

  1. Clone this repository. Set up a new virtual environment and activate it.
  2. Install dependencies using: 
    pip install -r requirements.txt

Usage

Activate your new virtual environment.

  1. Run prime_detection.py:

    python prime_detection.py
    • Follow prompts to select audio input devices.

  2. Adjust Detection Parameters:

    • Modify eps and min_samples in detect_and_cluster_bursts for DBSCAN clustering as needed.

Output

  • Hourly and daily burst counts logged to CSV files.
  • Annotated spectrogram images saved as spectrogram2detected.jpg.

Future Work

  • Optimize ORB and DBSCAN parameters for better burst detection.
  • Integrate real-time burst visualization.