Convolutional neural networks for charged particle detection using Hough transform in the ATLAS detector at HL-LHC

Due to increased luminosity and bunch crossing rate at the HL-LHC an improvement to the trigger system at the ATLAS detector is needed. In this thesis convolutional neural networks are suggested to improve charged particle candidate finding in Hough transformed images of hits in the Inner tracker. The results presented are based on simulated proton-proton collisions and single muons within a region 0.3 < φ0 < 0.5, 0.1 < η < 0.3. As an example at 99% the average candidate count is found to decrease from 216 to 97 by using a two-layer CNN method compared to the current. This is a significant decrease that could have a significant influence on the viability of the Hough transform method when considering the future trigger system for ATLAS at the LHC.