Neutron-gamma pulse shape discrimination for organic scintillation detector using 2D CNN based image classification.

Area of Science:

• Nuclear Physics
• Machine Learning
• Signal Processing

Background:

• Neutron-gamma pulse shape discrimination (PSD) is crucial for identifying different radiation types.
• Traditional methods like Charge Integration (CI) have limitations in feature extraction from raw signals.

Purpose of the Study:

• To implement and evaluate a novel 2D Convolutional Neural Network (CNN) for neutron-gamma PSD.
• To compare the performance of the 2D CNN approach against the conventional Charge Integration (CI) method.

Main Methods:

• Utilized a BC501A detector exposed to a Cf-252 source to collect neutron and gamma signals.
• Developed a data-driven approach for labeling digitized signals to create realistic training datasets.
• Applied a two-dimensional convolutional neural network to analyze unprocessed, digitized signal snapshots.

Main Results:

• The 2D CNN-based PSD approach achieved 99% accuracy on an independent dataset.
• The algorithm demonstrated comparable accuracy to the Charge Integration (CI) method.
• The CNN effectively extracts features directly from raw signal structures.

Conclusions:

• The proposed 2D CNN method offers a computationally efficient and accurate alternative for neutron-gamma PSD.
• The algorithm's ability to process raw signal data makes it suitable for diverse neutron detector applications.
• Data augmentation further enhances the robustness and applicability of the CNN-based PSD technique.