Zernike Moment Based Classification of Cosmic Ray Candidate Hits from CMOS Sensors

Area of Science:

• Astrophysics and Particle Physics
• Computer Science and Machine Learning

Background:

• Cosmic ray detection experiments utilizing CMOS technology require robust methods for artefact rejection and signal classification.
• Distinguishing between particle hits (spots, tracks, worms) and noise (artefacts) is crucial for data integrity.

Purpose of the Study:

• To evaluate the effectiveness of feature-based statistical classifiers for categorizing particle candidate hits in CMOS-based cosmic ray detectors.
• To propose an optimized feature extraction process using Zernike moments with preprocessing and denoising.

Main Methods:

• Utilized Zernike moments as feature descriptors for image data of particle candidate hits.
• Implemented and compared basic statistical classifiers and their ensemble extensions.
• Developed a preprocessing and denoising scheme to enhance feature extraction efficiency.

Main Results:

• Feature-based classifiers demonstrated a clear link between extracted features and geometrical properties of candidate hits.
• Ensemble extensions of the classifiers generally outperformed basic versions.
• Achieved an average recognition accuracy of 88% for classifying particle hits into four categories.

Conclusions:

• Feature-based statistical classifiers, particularly ensemble versions, are effective tools for artefact rejection and signal classification in CMOS-based cosmic ray detection.
• The proposed Zernike moment-based feature extraction with preprocessing offers an efficient approach for this task.
• These methods provide interpretable insights into particle hit characteristics, unlike black-box neural networks.