Convolutional Neural Network for Noise Reduction and Particle Tracking

Mikus, Andrey (2021) Convolutional Neural Network for Noise Reduction and Particle Tracking. Bachelor's Thesis, Physics.

Preview	Text bPHYS_2021_AndreyM.pdf Download (3MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (120kB)

Abstract

Thesis investigates the applications of a convolution neural network, CNN, for noise reduction and track identification of e+e− collision data taken from the drift chamber of BESIII. The CNN was originally developed and tested by Harmjan de Vries in his Master’s thesis. The noise reduction capabilities were further explored in a follow-up Bachelor’s thesis by Ignacio Gran ̃a. The network was shown to be effective in filtering noise and classifying individual particle tracks for Monte Carlo data simulated for e+e− collisions in the BESIII experiment. In this thesis, the CNN was trained and further tested for five different event topologies.. The network was successful in filtering noise from events containing up to 20 particle tracks. The feasibility of track identification, broken down into individual track labeling, charge classification and pair recognition was demonstrated. The effect of broken layers on the performance was also examined. The performance of the network for track identification decreases significantly for events with a large number of tracks. We observed that the network is particularly sensitive to the underlying features directly linked to the curvature of tracks, and, therefore, to the transverse momentum and electric charge of the particles. Tracks can be identified successfully in the case they are distinct in those features and among each other. For pair recognition, the opening angle between the tracks appears to be a key feature for discrimination.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Messchendorp, J.G. and Kavatsyuk, M.
Degree programme:	Physics
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	12 Jul 2021 08:44
Last Modified:	12 Jul 2021 08:44
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/25151

Actions (login required)

View Item