Classifying cosmic-ray proton and light groups in LHAASO-KM2A experiment with graph neural network

Precise measurement about the cosmic-ray (CR) component knee is essential for revealing the mistery of CR's acceleration and propagation mechanism, as well as exploring the new physics. However, classification about the CR components is a tough task especially for the groups with the atomic number close to each other. Realizing that the deep learning has achieved a remarkable breakthrough in many fields, we seek for leveraging this technology to improve the classification performance about the CR Proton and Light groups on the LHAASO-KM2A experiment. In this work, we propose a fused Graph Neural Network model in combination of the KM2A arrays, in which the activated detectors are structured into graphs. We find that the signal and background can be effectively discriminated in this model, and its performance outperforms both the traditional physics-based method and the CNN-based model across the whole energy range.