Optimization of the clustering algorithm of the CYGNO experiment

Abstract

In general, the task of clustering objects in images might be simple and several algorithms have been developed for this purpose. However, the performance of such algorithms needs to be understood in the specific environments of the application and, additionally, when it comes to the identification of rare events, with low signal-to-noise ratio, it becomes even more necessary to study and, eventually, optimization of these algorithms considering the particularities of the problem faced, as in the case of the CYGNO experiment that is developing a new detection system based on a TPC Triple GEM detector coupled to a low noise and high spatial resolution CMOS sensor. In this context, two of the most commonly mentioned clustering algorithms in the scientific literature known as Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Nearest Neighbor Clustering (NNC) were evaluated in the CYGNO environment. Using this study, this thesis work offers a proposal to adapt the DBSCAN algorithm, called intensity-DBSCAN (iDBSCAN), and makes a comparative study between the methods studied. A description of the iDBSCAN algorithm, including testing and validating its parameters, and a comparison with the DBSCAN itself and the NNC using data acquired with one of the CYGNO detector prototypes will be presented. The achieved results show that the adapted version of DBSCAN is capable of providing a detection efficiency as good as those obtained with the classical algorithms and, at the same time, improve the energy resolution and background rejection of the detector