Improving the machine learning based vertex reconstruction for large liquid scintillator detectors with multiple types of PMTs

TL;DR

This paper enhances machine learning methods for vertex reconstruction in large liquid scintillator detectors by optimizing input data, leading to significant improvements in resolution at different energies.

Contribution

It introduces an optimized input image approach that separates PMT types and includes second hit information, improving vertex resolution.

Findings

Vertex resolution improved by about 9.4% at 1 MeV

Vertex resolution improved by about 9.8% at 11 MeV

Enhanced input data significantly boosts reconstruction accuracy

Abstract

Precise vertex reconstruction is essential for large liquid scintillator detectors. A novel method based on machine learning has been successfully developed to reconstruct the event vertex in JUNO previously. In this paper, the performance of machine learning based vertex reconstruction is further improved by optimizing the input images of the neural networks. By separating the information of different types of PMTs as well as adding the information of the second hit of PMTs, the vertex resolution is improved by about 9.4 % at 1 MeV and 9.8 % at 11 MeV, respectively.