Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Application of convolutional neural networks for extensive air shower separation in the SPHERE-3 experiment

E. L. Entina, D. A. Podgrudkov, C. G. Azra, E. A. Bonvech, O. V. Cherkesova, D. V. Chernov, V. I. Galkin, V. A. Ivanov, T. A. Kolodkin, N. O. Ovcharenko, T. M. Roganova, M. D. Ziva

TL;DR

This work addresses the challenge of separating extensive air shower Cherenkov light from background in the SPHERE-3 detector by introducing a lightweight convolutional neural network as a second-stage trigger within a two-stage system. Using a detailed Monte Carlo pipeline, the authors generate a labeled dataset that mimics detector response and train a compact CNN to classify short, time-sequenced image patches from a 2653-pixel mosaic. The results demonstrate high baseline accuracy with a 1.0% false-positive rate which can be reduced to 0.1% at the cost of a 2.8% increase in missed events by applying a class-separation threshold, implying an effective energy threshold around $12$--$15$ PeV for SPHERE-3. The study shows the viability of CNN-based, real-time pattern recognition for reducing trigger rates while maintaining detection efficiency, with practical implications for setting SPHERE-3 energy sensitivity and guiding future hardware implementations.

Abstract

A new SPHERE-3 telescope is being developed for cosmic rays spectrum and mass composition studies in the 5--1000 PeV energy range. Registration of extensive air showers using reflected Cherenkov light method applied in the SPHERE detector series requires a good trigger system for accurate separation of events from the background produced by starlight and airglow photons reflected from the snow. Here we present the results of convolutional networks application for the classification of images obtained from Monte Carlo simulation of the detector. Detector response simulations include photons tracing through the optical system, silicon photomultiplier operation and electronics response and digitization process. The results are compared to the SPHERE-2 trigger system performance.

Application of convolutional neural networks for extensive air shower separation in the SPHERE-3 experiment

TL;DR

This work addresses the challenge of separating extensive air shower Cherenkov light from background in the SPHERE-3 detector by introducing a lightweight convolutional neural network as a second-stage trigger within a two-stage system. Using a detailed Monte Carlo pipeline, the authors generate a labeled dataset that mimics detector response and train a compact CNN to classify short, time-sequenced image patches from a 2653-pixel mosaic. The results demonstrate high baseline accuracy with a 1.0% false-positive rate which can be reduced to 0.1% at the cost of a 2.8% increase in missed events by applying a class-separation threshold, implying an effective energy threshold around -- PeV for SPHERE-3. The study shows the viability of CNN-based, real-time pattern recognition for reducing trigger rates while maintaining detection efficiency, with practical implications for setting SPHERE-3 energy sensitivity and guiding future hardware implementations.

Abstract

A new SPHERE-3 telescope is being developed for cosmic rays spectrum and mass composition studies in the 5--1000 PeV energy range. Registration of extensive air showers using reflected Cherenkov light method applied in the SPHERE detector series requires a good trigger system for accurate separation of events from the background produced by starlight and airglow photons reflected from the snow. Here we present the results of convolutional networks application for the classification of images obtained from Monte Carlo simulation of the detector. Detector response simulations include photons tracing through the optical system, silicon photomultiplier operation and electronics response and digitization process. The results are compared to the SPHERE-2 trigger system performance.
Paper Structure (7 sections, 1 equation, 5 figures, 5 tables)

This paper contains 7 sections, 1 equation, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Method
  4. Data Preparation
  5. Implementation details
  6. Results
  7. Discussion

Figures (5)

  • Figure 1: SPHERE-3 experiment scheme.
  • Figure 2: Example of a topologiсal trigger activation. On the right are shown all possible triplet variations (without rotations).
  • Figure 3: Example of an EAS CL structure on the mosaic from a primary 10 PeV iron nuclei. The image represents some instant signal values in pixels and not the total signal collected. The crescent shape of EAS is the result of photon arrival delays due to differences in optical path lengths.
  • Figure 4: Dataset preparation step. The red highlighted areas contain only noise, and the green highlighted area contains the EAS-event signal.
  • Figure 5: Distribution of the third pixel amplitude in the brightest triplet of the EAS image.