An algorithm for precise segmentation of the laser curve of train whee ls based on Encoder-Decoder network is proposed.In view of the rich local featu res and simple semantic features of the laser stripe images of train wheels,a n etwork with shallow depth,high resolution and good detail performance is design ed.The designed network makes good use of the dense link mechanism and up-sampl ing module,strengthens feature reuse and feature propagation,has fewer paramet ers and can extract contextual semantic information at multiple scales.Experime nts have proved that the Encoder-Decoder network shows excellent performance in train wheelset extraction compared to other networks.Based on the Encoder-Decod er network in the data set of the laser stripe laser curve of the train wheel se t,the intersection ratio,recall rate,accuracy and F1_score indicators reached 86.5%,89.2%,99.9%,85.0%,respectively,which can extract the train wheel set s more accurately.Laser stripes.At the same time,the Encoder-Decoder network can improve the effect of noise on the fringes to a certain extent when the trai n wheel pair laser fringe segmentation is performed.Therefore,it has a good ap plication prospect in railway safety.