Multiview CNN Architectures for Primary Particle Classification in Extensive Air Showers

Extensive air showers (EAS) from hadronic and gamma-ray origins are central to high-energy astrophysics, which has motivated the development of automated classification methods. This study explores the use of deep convolutional neural networks (CNNs) for EAS classification using CORSIKA 7.7550 simulations, which generate synthetic data on cosmic-ray cascades. Multiview images across the XZ, YZ, and XY planes, in both color and grayscale, were constructed to provide a robust training dataset. The proposed CNN framework was designed to enhance the separation between hadronic and gamma-induced events, supporting the accurate identification of primary particles, which is a key requirement for rapid characterization of high-energy cosmic phenomena.In this study, neural architectures (ResNet-50, Compact, and AlexNet) were systematically evaluated across different energies and zenith angles, considering particles such as iron, protons, positive muons, and gamma rays. The results indicate a strong performance in classifying heavy nuclei and muons, whereas distinguishing protons from gamma rays remains challenging because of the complex morphology of their cascades and atmospheric interactions. Among the tested models, Compact achieved an accuracy of up to 97.22% on grayscale images with a minimal inference time (1.19 ms), whereas ResNet-50 achieved an accuracy of 97.00 % using color images with data augmentation. These findings highlight the potential of CNN-based approaches for advancing automated EAS analysis in high-energy astrophysics.