Instance Segmentation for Neutrino Experiments using Deep Learning
Yu, Felix J.
2022
-
In recent years, deep learning has played an emerging role in event reconstruction for neutrino experiments using liquid argon time projection chambers (LArTPCs), a high-precision particle imaging technology. MicroBooNE is one such example of an experiment which utilizes deep learning to analyze its 2D output images. Given that these images are essentially an assortment of particle interactions, ... read morewe report on the effectiveness of object detection neural networks such as Mask Region-Convolutional Neural Network (Mask-RCNN) in analyzing these interactions. Mask-RCNN is widely used in computer vision problems and has three main goals: to identify the location of each object in an image using a bounding box, to classify an object in each bounding box, and to cluster each object by determining its pixel boundaries using a mask. Furthermore, we explore upgrading MaskRCNN to suit the specific needs of neutrino event data. A new architecture, Sparse Mask-RCNN (sMask-RCNN), is developed, a much more efficient version of the network to be used on neutrino event data due to the sparsity of the images. A 3D version of the network is also developed to satisfy the growing need for reconstruction algorithms that work on 3D image data for future experiments such as DUNE. Inspired by the conversion to 3D, we introduce a sparse bounding box proposal method that greatly reduces inefficiencies associated with box predictions in 3D. Lastly, we explore both the speed and performance capabilities of 3D Mask-RCNN on clustering electromagnetic shower and michel electron interactions in simulated 3D LArTPC data.
Thesis (B.S.E.P.)--Tufts University, 2022.
Submitted to the Dept. of Physics and Astronomy.read less - ID:
- 9019sh28d
- To Cite:
- TARC Citation Guide EndNote
- Usage:
- Detailed Rights
