An Application Study of Car Body Number Recognition Based on PGNet

Cargo transportation is an essential part of railroad transportation. The compartment number is a symbol to distinguish different freight carriages, and accurate identification of train car numbers plays a vital role in determining car weight information, settlement of transportation costs, and safe operation of railroad transportation. This article is based on the end-to-end text recognition algorithm PGNet (Point Gathering Network) in the PaddleOCR model built by Baidu Feijiang. Different training optimization strategies are adopted to apply and study the problem of box number recognition in railway and train transportation scenarios. The experimental results show that the carriage character recognition accuracy reaches a percent of 94.5, and has certain stability and robustness in the actual process of field deployment applications.