At present, transformer winding strain monitoring is divided mainly into off-line detection and on-line detection. Due to the interference of the complex electromagnetic environment, on-line detection has not been widely used. Although off-line detection is more mature, it can not accurately judge the winding strain form. Based on the above problems, this research investigated a strain gauge strain detection method based on distributed fiber optic sensing, and proposes a winding strain identification method based on the S-transform and an extreme learning machine (ELM). First, the deformation of the winding in the process of transformer operation is simulated, and the corresponding Brillouin frequency shift is collected. Then, the time-frequency analysis of the strain signal is carried out using an S-transform, and the transformed time-frequency feature is extracted as the input sample to the neural network. An ELM was used for training identification. Experimental results show that the method can effectively identify the common winding deformation form, and that the recognition effect is better and the accuracy is high.