Palestrante
Descrição
One of the most current problems in the detection of cosmic radiation by the Cherenkov Telescopes is the identification of the primary particle initiated by gammas, diffuse-gammas and electrons. The images generated in the cameras, by both these primaries, are very similar, making their identification not a simple task. In addition to the fact that gamma primaries can be diffuse sources. On the other hand, the images generated by primaries initiated by heavier particles, such as protons, helium and even iron, are different when compared to those of gamma; however, the quality of the generated image and the detection statistics are not accurate enough when using traditional approaches. This work proposes the use of images from these telescopes as input for machine learning to study the identification of these primary particles. Some initial studies showed the need to use images with more pixels for a better identification, so in this work we will only use larger telescopes (more than 20 meters diameter); in the Cherenkov Telescope Array (CTA) these are called LST (Larger Size Telescopes). Preliminary results showed up to 98% of accuracy for gammas identification and up to 83% for electrons, using 4 telescopes in stereo detection. A secondary objective that naturally shows after the first study is to verify if this same methodology could also be capable of associating this learning with macroscopic quantities. The network was redesigned for the reconstruction of the primary energy. For this reconstruction the correlation coefficient, between the simulated values and the predicted by the network, reached $R^2$ = 0.95 for a global fit. Thus, the main contributions for this paper are: the ability to classify the particle type and fit the Energy of the shower and to integrate different sources for this purpose.
