Impact of filtering on CYGNO experiment

Abstract

This work proposes to study the impact of classic digital filters and a convolutional neural network in the detection process of pixels with the presence of a signal formed from the release of energy produced by particles that interact inside a TPC detector which makes use of an optical readout based on high-resolution image sensors. For this analysis, an evaluation methodology based on the use of simulation data obtained within the scope of the CYGNO Experiment is also proposed. Finally, real data are analyzed as a way to validate the results. Some classical filtering techniques were selected for comparative purposes, together with a convolutional neural network trained to perform the selection of pixels of interest, with the objective of verifying the advantages of using such techniques in a pre-processing stage of the data, especially regarding energy estimation and processing time. The results obtained showed that a convolutional network has the potential to improve the performance of the processing stage of the images generated by the detector and that a classical non-linear filtering can reproduce a result similar to the algorithm used by the experiment in a time four times shorter.