Classificação de chuveiros atmosféricos utilizando redes neurais convolucionais

Abstract

Every second, countless particles from space reach our planet, some interacting with the Earth’s atmosphere, which acts as a shield. This is the case for cosmic rays and gamma rays, which generate a shower of lighter and less energetic particles as they penetrate the atmosphere. Due to the opacity of the atmosphere to gamma rays, Imaging Atmospheric Cherenkov Telescopes (IACTs) such as HESS, VERITAS, MAGIC, and the upcoming CTA, are essential for observing these events. A particular characteristic of gamma rays is that they travel through the Universe without the deflection suffered by charged particles, thus carrying information about the location where they were generated, allowing for detailed studies of these sources. This work explores the use of deep learning techniques, specifically convolutional neural networks, for analyzing data from these telescopes, separating events initiated by gamma rays from those initiated by hadrons. The training dataset was obtained through computational simulations using the CORSIKA software, generating cosmic showers initiated by protons and high-energy photons. Subsequently, the sim_telarray package was used to simulate the bunch of photons captured by the IACT cameras. These images were used to train a convolutional neural network with the objective of distinguishing showers initiated by gamma rays from hadronic ones. The model achieved an accuracy of over 90% in distinguishing between these two types of events, demonstrating its great potential applicability in the analysis of data from these kind of events.