Because the signal noise ratio (SNR) of the optoelectronic information in the Brillouin scattering spectrum of distributed sensing systems is too low, the measured inversion becomes very difficult. In order to extract features for Brillouin scattering spectrum of distributed sensing systems in high accuracy, a novel hybrid algorithm based on Levenberg-Marquardt (L-M) and particle swarm optimization (PSO) is proposed. PSO is applied to get a volume of gross unites which served as the initial value of L-M, then output the result of L-M as the final result. This algorithm overcomes the premature local extremum of PSO and excessive dependence for initial value of L-M. It can ensure the efficiency and precision of the value. Numerical results show that the algorithm can be used in different weight ratios, different line widths, low SNR and scattering spectrum parameter estimation in large scope. The error of the hybrid algorithm is 2.18838 MHz with SNR of 10 dB, which is better than the other two. The experimental results indicate that the new algorithm can extract features for Brillouin spectrum with different pulse widths. It can increase the accuracy of Brillouin scattering spectrum.