Classification of radioxenon spectra with deep learning algorithm.

Area of Science:

• Nuclear Engineering
• Data Science
• Spectroscopy

Background:

• Radianoble gas detection is crucial for nuclear event monitoring.
• Current methods for analyzing radioxenon spectra can be time-consuming.
• Automated classification is needed to improve efficiency and accuracy.

Purpose of the Study:

• To develop and validate a deep learning model for classifying Beta-Gamma coincidence radioxenon spectra.
• To assess the effectiveness of Convolutional Neural Network (CNN) techniques for spectral analysis.
• To establish a pre-screening method for nuclear event detection.

Main Methods:

• Utilized a Convolutional Neural Network (CNN) deep learning model.
• Trained the model on actual radioxenon spectral data from the USX75 station (2012-2019).
• Applied the model for classifying clean background and CTBT-relevant samples.

Main Results:

• Achieved high classification average accuracies: 92% for clean background and 98% for CTBT-relevant samples.
• Demonstrated the model's capability as an effective pre-screening tool.
• Showcased the successful integration of nuclear engineering principles with deep learning.

Conclusions:

• The proposed deep learning approach offers a promising method for accelerating and optimizing radioxenon spectral analysis.
• CNN-based classification can serve as a valuable pre-screening tool, reducing reliance on manual thresholding.
• Combining nuclear engineering expertise with AI enhances the detection and review of nuclear events.