The cryogenic refueling system is an important part of the launch site ground support equipment, includs the storage, transportation, supply, control and safety of cryogenic medium. Due to the low temperature boiling and volatile characteristics of cryogenic propellant, its refueling process is very complicatal. In order to meet the precise refueling requirements of the new generation of launch vehicle propellant, the liquid level in the tank needs to be monitored accurately during the refueling process in real time. This paper analyzes and extracts the characteristics of the triangular wave voltage and linear wave voltage for the liquid level signal data of the rocket ground refueling process, completes the identification and detection of different refueling states based on the BP neural network algorithm, and applies it to the sensor node discrimination, which can optimize the level calculation algorithm, reduce the demand for human intervention and improve the accuracy of liquid level measurement. The experimental results show that the method can identify the low temperature refueling state effectively, and the accuracy rate is over 90%, and can be used in liquid level signal processing to significantly improve the accuracy of level height calculation.