Based on the correlation method and the Levenberg-Marquardt (LM) algorithm, this study proposes a correlation-LM algorithm for fitting the central frequency of the Brillouin spectrum. The proposed algorithm exploits the correlation method because it does not require an initial value and is insensitive to noise. Therefore, the correlation method is used to determine the iterative initial value. Inputting the initial value into the LM algorithm for iteration yields less-accurate results. Accurate information of the Brillouin scattering spectrum is finally obtained. Thus, overcoming the problem of iteration using the LM method is necessary as it has the disadvantage of being extremely sensitive to the initial value. Therefore, this study also proposes a new bimodal Brillouin spectrum fitting method and the Brillouin peak selection principle with double peaks. This algorithm can effectively improve the fitting accuracy and the fitting efficiency, making the single-peak fitting error within 1 MHz, the double-peak fitting error within 2 MHz, and the fitting degree to be 0.9988. To test the practicability of the algorithm, it is used to detect the transformer winding deformation. Two strains are applied to the modified winding. The correlation-LM algorithm could extract the central frequency accurately and decouple the strain information. The parallel test error is within 3 MHz, the fitting degree reaches 0.9964, and the strain is accurately located, thereby proving the superiority and practicality of this method.