To the problem of the peak detection that could not be adaptively solved in the multi-peak fiber Bragg grating (FBG) signal, a self-adaptive multi-peak detection algorithm is proposed. This algorithm uses the sliding mean filtering method to remove the noise in spectral signal, and combines with the Hilbert transform to adaptively segment the peak area of the multi-peak spectrum. By analyzing the asymmetric characteristic of spectral peak, a peak value is compensated by the strategy based on the asymmetric generalized Gaussian model for improving position precision of spectral peak. Experimental results show that the proposed algorithm could gain higher accuracy and better stability than the comparing algorithms, and the detection error is under 1 pm. The proposed algorithm impacts on the multi-peak detection of distributed sensor networks.