Aiming at the problems of low accuracy and poor reliability of existing signal de-noising methods in the signal processing of fiber Bragg grating sensing system, based on the principle of fiber Bragg grating sensing, a signal de-noising method of fiber Bragg grating sensing system combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method and S-G filtering method is proposed. The CEEMDAN method is used to decompose the sensor system signal, perform S-G filtering on the decomposed intrinsic mode functions (IMF) component, and reconstruct the filtered IMF component. Through the comparative analysis of various noise containment conditions, the feasibility of the proposed method is verified. The results show that compared with the traditional de-noising method, the proposed method has better de-noising performance, the wavelength fluctuation is greatly reduced by more than 90%, and the signal-to-noise radio (SNR) and peak signal-to-noise radio (PSNR) indicators are increased by about 1dB, which can effectively reduce the signal noise under the premise of maintaining the detailed characteristics of the sensing signal, which has certain use value.