The baseline of the fiber Bragg grating (FBG) sensor network spectrum signal drifts, which is caused by the environmental conditions. To address this issue, this study proposes a spectrum baseline correction method based on the improved empirical mode decomposition (EMD). When this method is used to process simulation data, the spectrum after processing can be divided into the reflection peak regions of each FBG by setting the threshold value. The feasibility of the improved EMD method is verified. The spectral signals of the FBG sensor network with baseline drift under the influence of optical fiber path loss and end reflection are collected by experiments. The improved EMD method, the wavelet soft threshold method and the penalty least square method are used to achieve the real-time correction of the spectrum. The Gaussian nonlinear fitting method calculates the center wavelength of each FBG of the normal and corrected spectra. Results show that the survival rates of the sensing signal are increased by 48. 9% and 61. 6% , with the average deviation and deviation of each FBG center wavelength calculated after the use of the improved EMD method. The standard deviation values are the smallest, which are 4. 33, 6. 28 and 6. 01, 6. 58 pm. The spectral signal corrected by this method could provide reliable information for physical quantity detection.