Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Selection of gamma events from IACT images with deep learning methods

E. O. Gres, A. P. Kryukov, A. P. Demichev, J. J. Dubenskaya, S. P. Polyakov, A. A. Vlaskina, D. P. Zhurov

TL;DR

This work addresses gamma/hadron separation in TAIGA-IACT data by incorporating wobbling pointing mode into deep-learning analysis. It compares image-based CNNs, Hillas-parameter networks, and a hybrid approach using Monte Carlo TAIGA-IACT simulations, augmented to reflect ON/OFF wobble conditions. The results show that wobbling augmentation is crucial for reliable classification and that CNNs and Hillas-parameter networks offer comparable performance, achieving substantial hadron suppression but still struggling to detect gamma signals under severe class imbalance. The study highlights the need for more data, refined preprocessing, and hyperparameter tuning to realize practical gamma-ray detection with deep learning in TAIGA-IACT data.

Abstract

Imaging Atmospheric Cherenkov Telescopes (IACTs) of gamma ray observatory TAIGA detect the Extesnive Air Showers (EASs) originating from the cosmic or gamma rays interactions with the atmosphere. Thereby, telescopes obtain images of the EASs. The ability to segregate gamma rays images from the hadronic cosmic ray background is one of the main features of this type of detectors. However, in actual IACT observations simultaneous observation of the background and the source of gamma ray is needed. This observation mode (called wobbling) modifies images of events, which affects the quality of selection by neural networks. Thus, in this work, the results of the application of neural networks (NN) for image classification task on Monte Carlo (MC) images of TAIGA-IACTs are presented. The wobbling mode is considered together with the image adaptation for adequate analysis by NNs. Simultaneously, we explore several neural network structures that classify events both directly from images or through Hillas parameters extracted from images. In addition, by employing NNs, MC simulation data are used to evaluate the quality of the segregation of rare gamma events with the account of all necessary image modifications.

Selection of gamma events from IACT images with deep learning methods

TL;DR

This work addresses gamma/hadron separation in TAIGA-IACT data by incorporating wobbling pointing mode into deep-learning analysis. It compares image-based CNNs, Hillas-parameter networks, and a hybrid approach using Monte Carlo TAIGA-IACT simulations, augmented to reflect ON/OFF wobble conditions. The results show that wobbling augmentation is crucial for reliable classification and that CNNs and Hillas-parameter networks offer comparable performance, achieving substantial hadron suppression but still struggling to detect gamma signals under severe class imbalance. The study highlights the need for more data, refined preprocessing, and hyperparameter tuning to realize practical gamma-ray detection with deep learning in TAIGA-IACT data.

Abstract

Imaging Atmospheric Cherenkov Telescopes (IACTs) of gamma ray observatory TAIGA detect the Extesnive Air Showers (EASs) originating from the cosmic or gamma rays interactions with the atmosphere. Thereby, telescopes obtain images of the EASs. The ability to segregate gamma rays images from the hadronic cosmic ray background is one of the main features of this type of detectors. However, in actual IACT observations simultaneous observation of the background and the source of gamma ray is needed. This observation mode (called wobbling) modifies images of events, which affects the quality of selection by neural networks. Thus, in this work, the results of the application of neural networks (NN) for image classification task on Monte Carlo (MC) images of TAIGA-IACTs are presented. The wobbling mode is considered together with the image adaptation for adequate analysis by NNs. Simultaneously, we explore several neural network structures that classify events both directly from images or through Hillas parameters extracted from images. In addition, by employing NNs, MC simulation data are used to evaluate the quality of the segregation of rare gamma events with the account of all necessary image modifications.
Paper Structure (8 sections, 4 equations, 3 figures, 2 tables)

This paper contains 8 sections, 4 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Wobbling Augmentation for IACT Images
  3. Neural Networks for TAIGA-IACT Data Classification
  4. Training samples
  5. Considered NN architectures
  6. Validation results
  7. Test results
  8. Conclusion

Figures (3)

  • Figure 1: Scheme of the proposed wobbling augmentation necessary to standardize image transformations between experimental data and a neural network’s standard image form.
  • Figure 2: Architecture of neural networks considered in the work in the task of classification of IACT images with wobbling augmentation: a) Convolutional neural network predicting directly from IACT images; b) Fully connected neural network, predicting images' class from Hillas parameters calculated previously from the images; c) Combined convolutional neural network predicting from both images and Hillas parameters.
  • Figure 3: Distributions of NN predictions for gamma and proton events in case of ON (a) and OFF (b) pointing mode. The results of convolutional neural network analyzing only images are presented. Distributions for other NNs differ little from that shown