Feature extraction and classification of calorimeter data with neural nets

Several neural-net approaches for classification tasks on calorimeter data are presented. Only the energies deposited in the calorimeter cells are taken into account. The simple method consists of using these energies as input into a classification net. Its performance is compared to two more sophisticated approaches where features are used for classification. They are either calculated using Zernike polynomials or extracted by special nets. thods are applied to the electron-pion separation for test-beam data of the RD1 Collaboration at CERN. Excellent results are obtained in comparison with the conventional approach. The simple method turns out to be the most efficient one.