Track irregularity is the main source of train vibration, which has a direct impact on the safety of train operation and passengers’comfort. On the basis of measured track irregularities, combined with the structural characteristics of high-speed maglev TR08 vehicle and the basic principle of deep neural network, TensorFlow neural network is used to characterize the relationship between track irregularities and vehicle vibration accele ration. In this paper, a method of detecting irregularities by measuring vibration acceleration and constructing neural network is proposed. The research results show that the relative accuracy of the predicted track irregularity value and the real value by the depth neural network is more than 99%, and it can measure the height and level irregularity at the same time, which provides a theoretical basis for the new method of track irregularity measurement.