Abstract: Pi2 pulsations are irregular ultra-low frequency waves, representing a significant transient response to the coupling between the magnetosphere and ionosphere. Their occurrence is associated with onset of substorms. As a disturbance phenomenon in the Earth’s magnetosphere, the occurrence signal of Pi2 pulsations is hidden within the observation data of geomagnetic field components. Addressing the increasing amount of observation data, how to efficiently determine whether Pi2 pulsation has occurred in a segment of geomagnetic field component observational data is the key to build a Pi2 pulsation identification model. Based on the time series observation data of the FGM from the Chinese Meridian Project and One-Dimensional Residual Convolutional Neural Network (1D-ResCNN), this paper establishes an end-to-end Pi2 pulsation identification model. This model can distinguish whether Pi2 pulsation occurs in the observation data of a certain geomagnetic field component. Experimental results show that this model has higher recognition accuracy and lower false positive rate and false negative rate than the existing Pi2 pulsation machine learning identification model.