Neural Networks for Particle Reconstruction in High Energy Physics Detectors

The core of the Particle Flow (PFlow) algorithms in High Energy Physics experiments lies in reconstructing the intrinsic nature and kinematic properties of both charged and neutral particles, utilizing the data provided by detectors. Through effective trajectory analysis methods for various particle types, it becomes evident that the magnetic field significantly influences the accuracy of final reconstruction efforts. In this paper, we present several computer vision models aimed at enhancing PFlow algorithms. We employ a Multilayer Perceptron (MLP) to address scenarios where no magnetic forces interfere with particle tracks. Subsequently, we conduct a comprehensive discussion and comparison of the experimental results obtained from various enhanced Convolutional Neural Network (CNN) models, such as Polar Coordinate Processing, Parameterized Differential Operators (PDOs), Cylindrically Sliding Windows (CSW), tailored to the characteristics of our datasets.