Improving the detection efficiency of IRan ANtineutrino Detector (IRAND) based on Convolutional Neural Network (CNN)

Abstract

Discriminating reactor electron antineutrino from the cosmic muon is a fundamental challenge in antineutrino detection using Inverse Beta Decay (IBD) interaction in scintillator-based detectors. This study introduces an innovative approach by employing a Convolutional Neural Network (CNN) in the IRAND-Sim-02 simulation package to distinguish between muon and antineutrino in the IRAND (a 10 × 10 array of segmented plastic scintillator IRan ANtineutrino Detector) for improving IRAND's performance. A training dataset was generated using Geant4 via the IRAND-Sim-02 package (analytical images-Temporal Probability Density, Time vs. Photocathode ID, and photocathode activation topology-to encode the spatiotemporal patterns of the events). For model evaluation, an independent test dataset, reflecting the true antineutrino-to-cosmic-muon ratio, was automatically produced by establishing a real-time Geant4-Python TCP socket connection. The overall accuracy of 99.97 % by CNN model demonstrates the precise event classification. Compared with traditional machine learning approaches, such as Artificial Neural Network (ANN) and Decision Tree models, the proposed CNN demonstrates greater reliability in distinguishing IBD events from cosmic muon backgrounds, particularly under class imbalance conditions, which highlights the considerable potential of our method for improving nuclear reactor monitoring and represent a significant advancement in the application of deep learning to fundamental particle detection.