In this paper, high-precision normal mode detection is achieved by using the observation data of the superconducting gravimeter and combining the normal time-frequency transform (NTFT) and optimal sequence estimation (OSE) methods. Firstly, the instantaneous phases of each station were extracted through NTFT, and it was found that the instantaneous phases of each station were not the same. It was speculated that there was a non-synchronous oscillation phenomenon in the normal mode. When different quantities or different combinations of station data were used, the detection results of the OSE method showed significant differences, indicating that this phenomenon had an important impact on the detection accuracy of the OSE method. Subsequently, OSE experiments were conducted using stations with relatively consistent phases, and it was found that the detection frequency was close to the PREM theoretical value, indicating that the issue of phase consistency must be taken into account when applying the OSE method. Based on this, this paper proposes a high-precision normal mode detection method that integrates NTFT and OSE and takes into account phase consistency. Taking the split state detection of the ₀S₂ model as an example, the effectiveness of this method was verified, and it was also demonstrated that phase consistency is a key factor for the effective application of the date stacking method. This research provides a reliable data foundation and technical support for deeply revealing the deep structure of the Earth and optimizing the Earth model.