Railway freight train number detection method based on deep learning
Xiucai Guo, Cheng Gongli · 2020
In view of the complicated environmental background of freight trains loading stations in coal mines, the accuracy of train number positioning is easily affected by various influences. This paper aims to optimize the EAST natural text detection model and apply it to the train number positioning of railway freight trains. The improved ResNet50 network is used to extract the car number feature, and the BLSTM is added to the Feature that is about to be deconvolved in each layer. At the same time, when the polygon is made, the shrink distance of 0.3 times is changed to 0.1 times. Experiments show that the vehicle number area can be accurately detected through optimization.