The traditional system-level calibration method regards the random noise caused by the thermostat compressor, turntable motor and other environments as white noise, but the output signal in the actual process has non-stationary and nonlinear characteristics, while the least squares fitting method is a linear regression, the error coefficient solved by this method becomes one of the key factors restricting the high precision of the fiber optic strap-down inertial navigation system. Aiming at the above-mentioned problems, an improved SINS system-level calibration algorithm based on Hilbert-Huang transform is proposed. The high-frequency random noise in the original output signal of the inertial instrument is removed based on Hilbert-Huang transform, and then the IMU error parameters are calculated by the system-level calibration improved method. Through multiple sets of experiments, it has been proved that this method can effectively improve the accuracy of IMU error parameter identification, and the dynamic navigation position error of 1 h is relatively reduced by about 10%.