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

2024-10-03

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.

Instrumentation and Methods for Astrophysics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively separate the signal generated by extensive air showers (EAS) from the background noise in the SPHERE - 3 experiment. Specifically, the SPHERE - 3 experiment uses the reflected Cherenkov light method to detect cosmic rays, but this method faces the problem of background noise from the reflection of starlight and airglow photons. In order to accurately record EAS events and reduce the false - trigger rate, an efficient trigger system is required.

### Specific description of the problem

1. **Background noise problem**:
- The SPHERE - 3 experiment studies cosmic rays by detecting Cherenkov light reflected from snow. However, starlight and airglow photons will also be reflected into the detector, forming background noise.
- This background noise will interfere with the accurate identification of EAS events, leading to false triggers.

2. **Limitations of the existing trigger system**:
- The topological trigger system used in SPHERE - 2 judges whether to trigger event recording based on the adjacent relationship between pixels. However, due to the optical crosstalk characteristics of SiPM (Silicon Photomultiplier), this method is no longer applicable in SPHERE - 3.
- Compared with the traditional PMT (Photomultiplier Tube), SiPM is smaller, lighter, and has lower power consumption, but its high sensitivity and optical crosstalk will lead to more false triggers.

3. **Objective**:
- Develop a new trigger system that can effectively distinguish EAS signals from background noise, thereby improving the detection efficiency and reducing the false - trigger rate.

### Solution

The paper proposes a two - stage trigger system based on convolutional neural network (CNN):

1. **First stage: Topological trigger**:
- Initially screen out pixel combinations that may contain EAS signals through topological logic.

2. **Second stage: Deep - learning classification**:
- Use CNN to further analyze the data screened out in the first stage and judge whether it is a real EAS event.
- CNN can recognize complex visual patterns, thereby effectively distinguishing signals from noise.

### Key formulas

- **Optical crosstalk probability**:
\[
f(n)=p^{n}F
\]
where \( p \) is the optical crosstalk probability, \( F \) is the background photon flux, and \( n \) is the number of crosstalk multiplicity times.

- **Random trigger rate**:
\[
N_{p}f_{t}^{k}\tau^{k - 1}
\]
where \( N_{p} \) is the number of pixels, \( f_{t} \) is the target activation rate, \( \tau \) is the pixel activation state duration, and \( k \) is the number of simultaneously activated pixels.

Through this method, the paper successfully solves the background noise separation problem in the SPHERE - 3 experiment and significantly reduces the false - trigger rate.