Muon reconstruction with a convolutional neural network in the JUNO detector

Abstract

The Jiangmen Underground Neutrino Observatory (JUNO) is designed to determine the neutrino mass ordering and measure neutrino oscillation parameters. A precise muon reconstruction is crucial to reduce one of the major backgrounds induced by cosmic muons. This article proposes a novel muon reconstruction method based on convolutional neural network (CNN) models. In this method, the track information reconstructed by the top tracker is used for network training. The training dataset is augmented by applying a rotation to muon tracks to compensate for the limited angular coverage of the top tracker. The muon reconstruction with the CNN model can produce unbiased tracks with performance that spatial resolution is better than 10 cm and angular resolution is better than 0.6 degrees. By using a GPU accelerated implementation a speedup factor of 100 compared to existing CPU techniques has been demonstrated.

Reconstruction of Muon Bundle in the JUNO Central Detector

“…15, and it is about 98 ms/event. The CPU implementation of the reconstruction method in this paper provides a huge processing speed improvement of a factor of 51 compared to the fastest light method (5000 ms/event) [6,7]. This speed makes it possible to apply the algorithm in the online event classification, which requires a fast online-filtering for a possible data rate reduction.…”

“…2 Schematic of a spherical target detector and different kinds of muon events labelled with single, bundle, through-going, stopping, and clipping. Three reconstruction algorithms have been developed for single muon, such as a method with a geometrical model which utilizes the geometrical shape of the fastest light [5], a method with the fastest light model which utilizes the minimization of the first hit time (FHT) [6], and new technology using deep learning and GPU acceleration [7]. But the reconstruction of muon bundles is not included and rarely discussed.…”

“…(6) Charge weighted calculation to correct each cluster seed and discard the fake seeds. (7) Merge the adjacent candidate seeds if they are too close. (8) Match two candidate seeds as a muon track.…”