Abstract: The problem of the identification for the locomotive number and types in the marshalling station always restricted the development of the integrated control technology of the locomotive.In order to solve this problem, this article studied the automatic identification of locomotive types and locomotive number, proposed an improved identification algorithm based on convolutional neural network LeNet-5, collected a lot of image materials of locomotive number. After image preprocessing, the training set was used to train the model, and a network model suitable for locomotive number identification was formed. The design of locomotive number identification system was implemented by using Python language and. NET platform. Experiments show that the method has reached a high recognition level for locomotive number identification. At present, the locomotive number identification system has been tested in Xiangyang North Station of Wuhan Railway Administration. The high-definition image materials are obtained from the high-definition cargo inspection system of the station, and the recognition effect is good, which provides strong technical support for the implementation of intelligent marshalling station.