In traditional sensing, each parameter is treated as a real number in the signal demodulation, whereas the electric field of light is a complex number. The real and imaginary parts obey the Kramers–Kronig relationship, which is expected to help further enhance sensing precision. We propose a self-Bayesian estimate of the method, aiming at reducing measurement variance. This method utilizes the intensity and phase of the parameter to be measured, achieving statistical optimization of the estimated value through Bayesian inference, effectively reducing the measurement variance. To demonstrate the effectiveness of this method, we adopted an optical fiber heterodyne interference sensing vibration measurement system. The experimental results show that the signal-to-noise ratio is effectively improved within the frequency range of 200 to 500 kHz. Moreover, it is believed that the self-Bayesian estimation method holds broad application prospects in various types of optical sensing.