Desenvolvimento de algoritmo de Trigger para detecção de sinais no experimento CYGNO

Abstract

The development of trigger algorithms with fast processing times is essential for experiments that process large volumes of data and seek rare events with a high acquisition rate. The CYGNO experiment, designed for the direct detection of dark matter through a gas detector, faces the challenge of processing approximately 80,000 images daily, many of which contain only electronic noise generated by the camera and detector. To address this challenge, this work proposes two trigger algorithms: one algorithm uses filtering with standard filters such as mean, median, and gaussian, as well as the matched filter; the other is based on convolutional neural networks (CNNs). Both methods achieved a detection rate of approximately 80% for 0.25 keV signals, with the filtering algorithm yielding a false alarm rate of 10%, while the CNNs reduced this rate to 1%. The proposed algorithms were able to detect all signals identified by the experiment’s reconstruction algorithm while operating with significantly lower processing times: the Gaussian filter took around 200 milliseconds on a CPU and 20 milliseconds on a GPU, whereas the CNN required approximately 550 milliseconds on a CPU and 200 milliseconds on a GPU. These advancements not only improve the efficiency of data analysis but also optimize the resources of the CYGNO experiment, enabling more effective management of the acquired images.