Impacts of Data Pre-processing on Jet Images Classification using Deep Learning

Jet images are essential utilities in High Energy Physics (HEP) as they enable the extraction of crucial physics information from particle collisions. Jets are clusters of particles produced during high-energy collisions. Jet images are two-dimensional representations of these jets that preserve important details about the energy distribution and the jet substructure. When it comes to the classification of different types of jets, deep neural networks yield very promising results. Convolutional Neural Networks (CNNs) have proven to be particularly effective in achieving highly accurate classifications. One common process in deep learning is data pre-processing. Data pre-processing involves techniques such as (background) noise reduction, image normalization, and feature extraction. The performance of deep learning models can be significantly impacted by the quality of pre-processed data. Therefore, this study discusses the significance, impact and usefulness of data pre-processing techniques on the jet images classification performance of deep learning models. This study makes use of reconstructed jet images of proton-proton collisions at the CERN Large Hadron Collider (LHC). The signal events are jets at final states produced in top anti-top events, whereas the considered background contributions that mimic the signal events are jets that yield in QCD events.