This report describes a new Convolutional Neural Network (CNN) model developed for the identification of gamma rays with the calorimeter of the DArk Matter Particle Explorer (DAMPE) mission. Its architecture is optimized in order to enhance the gamma ray/proton separation in the Monte-Carlo (MC) simulation data. Then, the classification performance is analyzed and compared with previously used methods. It is shown that this method significantly outperforms all the existing algorithms, both in gamma-ray efficiency and proton rejection. Finally, the model is successfully tested with the flight data in order to validate its use for further data analyses.