When we construct a large distributed sensing network by fiber Bragg grating (FBG), under the condition of same bandwidth, we fabricate as many fiber gratings as possible to increase the number of sensors, which leads to overlapping spectra and makes FBG central wavelength recognition difficult and reduces the demodulation accuracy. In order to solve this problem, we propose an improved particle swarm optimization (PSO) algorithm to improve the recognition accuracy of the central wavelength. First, the overlapping spectra model is established by spectral shape multiplexing technology. Then, the temperature experiment system is built to get the overlapping spectral signals, and the weight factor and the learning factor in the PSO algorithm are improved. Finally, the proposed algorithm is used to optimize the parameters of the overlapping spectra model, and it is compared with the six optimization algorithms. Simulation result and experimental result show that the proposed algorithm has the characteristics of fast convergence speed, short running time and high wavelength recognition accuracy compared with the contrast algorithm, and the wavelength demodulation error is less than 1 pm, which verifies the effectiveness and feasibility of the algorithm.