Based on the study of the sensing property of fiber bragg grating (FBG) sensor network, this paper takes hybrid wavelength division/time division multiplexing sensor network with high-capacity FBG as the research object to study noise source, noise category and traditional noise reduction algorithm. In accordance with the characteristic of crosstalk accumulation noise in the network, this paper proposes LCEEMD-LWT signal processing method, which uses the local complementary ensemble empirical mean decomposition (LCEEMD) method for signal preprocessing, and then adopts LWT technology to refine high-frequency signal and solve the related problems. Then this paper put forwards detrended fluctuation analysis (DFA) to evaluate the sensing signal of high-capacity FBG, preprocesses the sensing signal by LCEEWD-LWT method, and conducts a temperature sensing experiment. The average temperature error of the demodulating system is reduced from 0.2923℃ to 0.2357℃, which not only overcomes the shortcoming that the traditional signal processing method is not meticulous for high-frequency signal processing, but also avoids the problem that the signal processing time is too long, exerting a good effect on the signal pre-processing of high-capacity FBG. The method in this paper has certain theoretical and technical reference value for the signal processing and the improvement of sensing property of high-capacity FBG sensing demodulating system in engineering application.