Neural network applications for improved treatment of the EXCHARM experiment

The charged particles track recognition method based on Denby–Peterson segment model (DPSM) for Hopfield full-connected artificial neural network (ANN) is developed for handling of the EXCHARM experimental data. The specifics of the EXCHARM experiment (heavy background conditions, effects related to inefficiency of chambers and presence of secondary vertices) required the essential modification of the DPSM. The results of testing show that our modified ANN scheme has higher recognition efficiency than the current version of EXCHARM data processing software, but yields it in speed. The basic difference between two algorithms results in a small intersection of sets of badly recognized events. It gave us a possibility to create a combined event reconstruction algorithm based on the both current data processing program for majority of events and the ANN program for more complicated events. The combined approach allows to achieve 99% of event recognition efficiency in real conditions.