Semantic Segmentation with a Sparse Convolutional Neural Network for Event Reconstruction in MicroBooNE

TL;DR

This paper introduces SparseSSNet, a sparse convolutional neural network for pixel-level classification of MicroBooNE liquid argon TPC data, improving accuracy and efficiency in neutrino event analysis.

Contribution

The paper presents SparseSSNet, a novel sparse CNN that enhances semantic segmentation in liquid argon TPC data, marking its first application in this context and outperforming previous methods.

Findings

Achieves ≥99% accuracy on test data.

Processes an image in approximately 0.5 seconds.

Uses around 1 GB of memory, enabling CPU-based processing.

Abstract

We present the performance of a semantic segmentation network, SparseSSNet, that provides pixel-level classification of MicroBooNE data. The MicroBooNE experiment employs a liquid argon time projection chamber for the study of neutrino properties and interactions. SparseSSNet is a submanifold sparse convolutional neural network, which provides the initial machine learning based algorithm utilized in one of MicroBooNE's $\nu_e$-appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are re-classified into two classes more relevant to the current analysis. The output of SparseSSNet is a key input in further analysis steps. This technique, used for the first time in liquid argon time projection chambers data and is an improvement compared to a previously used convolutional neural network, both in accuracy and computing resource utilization. The accuracy achieved on the test sample is $\geq 99\%$. For full neutrino interaction simulations, the time for processing one image is $\approx$ 0.5 sec, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine.