Neutron-Gamma Pulse Shape Discrimination for Organic Scintillation Detector using 2D CNN based Image Classification

TL;DR

This paper presents a 2D CNN-based method for neutron-gamma pulse shape discrimination using raw detector signals, achieving high accuracy and improved efficiency over traditional methods.

Contribution

Introduces a novel 2D CNN approach for PSD that processes raw signals directly, offering a more efficient alternative to charge integration methods.

Findings

Achieved 99% accuracy on independent datasets.

Successfully differentiated neutrons and gammas with accuracy comparable to traditional methods.

Demonstrated the method's applicability to other neutron detectors.

Abstract

This study shows an implementation of neutron-gamma pulse shape discrimination (PSD) using a two-dimensional convolutional neural network. The inputs to the network are snapshots of the unprocessed, digitized signals from a BC501A detector. By exposing a BC501A detector to a Cf-252 source, neutron and gamma signals were collected to create a training dataset. The realistic datasets were created using a data-driven approach for labeling the digitized signals, having classified snapshots of neutron and gamma pulses. Our algorithm was able to successfully differentiate neutrons and gammas with similar accuracy as the CI approach. Additionally, the independent dataset accuracy for our suggested 2D CNN-based PSD approach is 99%. In contrast to the traditional charge integration method, our suggested algorithm with data augmentation, is capable of extracting features from snapshots of the raw data based on the signal structures, making it computationally more efficient and also appropriate for other types of neutron detectors.