Identification of e, μ, π by neural network in BES

Two different kinds of neural network methods, BP and LVQ, are applied to the identification of e, μ, π particles in BES experiment, and better results are obtained. The μ data samples used for training and tests are from cosmic events, e and π are strictly selected from real experimental data by very tight cuts. Although their momentum spectrums are non-uniform, interesting enough is that the selection efficiencies given by BP test results are quite uniform in the whole momentum range. The results from LVQ is little worse. It shows that BP is more powerful in pattern recognition than LVQ at least in this study