CNN-Based Classifier as an Offline Trigger for the CREDO Experiment.

Area of Science:

• Citizen science
• Particle physics
• Machine learning

Background:

• Citizen science projects like the Cosmic Ray Extremely Distributed Observatory (CREDO) use gamification to boost user engagement.
• Gamification, while beneficial, leads to an increase in data artefacts, including false signals and cheating.
• The CREDO Detector app collects cosmic ray data globally using smartphones.

Purpose of the Study:

• To develop a machine learning method for automatically tagging artefacts in the CREDO database.
• To distinguish between genuine cosmic ray signals and artefacts introduced by gamification.
• To reduce the manual effort required for data quality control.

Main Methods:

• Training a Convolutional Neural Network (CNN) to recognize morphological differences between signals and artefacts.
• Employing adaptive thresholding and Daubechies wavelet transforms to enhance image signal features.
• Validating the CNN's performance against human annotator assignments.

Main Results:

• The developed CNN-based trigger achieved a recognition ratio of almost 99% for both signals and artefacts.
• Wavelet transforms effectively amplified distinctive image features, improving recognition accuracy.
• The method successfully mimics human annotators' signal vs. artefact classifications.

Conclusions:

• The proposed machine learning approach provides an effective solution for artefact detection in citizen science data.
• This method significantly reduces the need for manual supervision in gamified citizen science experiments.
• The study demonstrates the potential of advanced signal processing techniques combined with CNNs for data quality assurance.