Aiming at the noisy signals collected by distributed optical fiber sensing systems, a noise reduction method to modified ensemble local mean decomposition (MELMD) combined with independent component analysis (ICA) is proposed, and the permutation entropy decision mechanism is introduced to improve the ability to suppress modal aliasing and spurious components. Firstly, the product function (PF) is obtained by decomposing the noisy signal by MELMD method, and the signal reconstruction is carried out. The difference between the noisy signal and the reconstructed signal is obtained to obtain the virtual noise, and the virtual channel is constructed. Then ICA is used to separate the signal-to-noise of the noisy signal and the virtual channel to obtain the final result. Experimental data were used to verify that the method was compared with EMD-ICA, EEMD-ICA, and MELMD by using indicators such as signal-to-noise ratio, mean absolute percentage error, and correlation coefficient as evaluation parameters. The results show that the proposed method can better eliminate the noise in the signal, retain the characteristic information of the signal, and provide a basis for the feature extraction and classification of optical fiber signal.