Optimization strategies for CNN's aiming crosstalk mitigation in finely segmented calorimetry systems

Calorimeters with fine sensor segmentation used in High Energy Physics experiments suffer from mutual interference due to electromagnetic coupling between neighboring sensor cells. Crosstalk affects the characterization and measurement of incident particle properties, like energy and time of flight, degrading the quality of event reconstruction. This work proposes different optimization strategies for a convolutional neural network using simulation-generated events, aiming to mitigate crosstalk effects. Using a strategy with per cell normalization, it was possible to enhance the timing estimation for the central cell, with a mean R² of 0.893, in comparison with 0.721 from the baseline.