This study aims to optimize the discrimination performance of gamma/proton in high altitude detection of astrological radiation (HADAR) experiments by employing the multilayer perceptron (MLP) algorithm. The HADAR experiment is used to observe the Cherenkov light produced by cosmic rays and gamma rays in the atmosphere via a composite array composed of four water lenses and surrounding scintillation detectors. And it is highly competitive in detecting the transient sources and the prompt emission of gamma-ray bursts due to its advantages such as low threshold energy (~30GeV) and wide field of view (~30°). However, the image discrimination between background noise and signal become weak when the detected energy of HADAR is less than 100 GeV, leading to unsatisfactory gamma/proton discrimination performance of traditional Hillas parameter methods. In this study, we employ MLP as a discriminator to conduct training and classification based on input characteristic parameters (such as Hillas parameters and core information). The results of Monte Carlo simulation demonstrate that the MLP method exhibits excellent performance and accuracy gamma/proton identification. Specifically, the discrimination between signal and background noise is enhanced at detection energies between 30-100 GeV, and the highest achieved Q-factor is 2.17 (proton exclusion rate ~97.80%, gamma retention rate ~32.20%). This study provides valuable references and a solid foundation for enhancing the gamma/proton discrimination performance of HADAR.