LArTPC Reconstruction with Deep Convolutional Neural Networks

Liquid argon time projection chambers (LArTPCs) are a type of particle detector widely used in current and future neutrino experiments. Their widespread use is due to (1) their ability to scale up to very large sizes and (2) their ability to take high-resolution images of the trajectories of charged particles that travel through them. These detectors will play a central role in advancing our knowledge of neutrinos and the fundamental laws of physics.

Example of LArTPC image from the MicroBooNE detector

In order to realize the discoveries our field hopes to make with LArTPC experiments, the information from LArTPC images must be used to reconstruct the true trajectories of particles as accurately as possible. These particle trajectories can be then be used to investigate potentially novel properties of the neutrino. Our group is heavily involved in advancing the application of deep convolutional neural networks to assist in the 3D reconstruction of particle interactions.

Goal of reconstruction: take 2D images and reconstruction 3D trajectories of particles. Image is from real data using tools underdevelopment

Example Applications

Here is a list of applications we have or are currently developing.

Reconstruction of space points

LArMatch description coming.

Identification of Keypoints

Keypoint network coming.

Classification of pixels/space points by trajectory type

SSNet coming.

Clustering of pixels into cosmic and neutrino interactions

SMask-RCNN coming.

Inferring missing trajectory segments in unresponsive channels

Infill coming.

Generative Networks

Generative networks work coming.

Classification of particle types

Classifier description coming.