Abstract: The traditional trouble of moving freight car detection system(TFDS) complete vehicle status inspection and analysis through image acquisition and manual interpretation, which is time-consuming and laborious. In recent years, with the development of artificial intelligence, intelligent recognition algorithms have gradually been added to TFDS systems. However, due to limited sample sizes and complex environments, the recognition accuracy of intelligent recognition algorithms still cannot meet the requirements of engineering applications. An intelligent monitoring and recognition method for the appearance status of coal transport trains based on feedback correction and weak light image enhancement is proposed in this paper, achieving efficient and high-precision intelligent recognition of vehicle appearance status. Firstly, to address the issue of low light image quality affecting the accuracy of intelligent recognition algorithms caused by constantly changing outdoor lighting conditions, a comparative study is conducted on the preprocessing performance of six low light enhancement algorithms on the original images and their impact on the accuracy of train carriage target recognition. The results show that after enhancement using the URetinex-Net method, the target recognition accuracy of train carriages can be improved from 55% to 90%. In addition, in order to solve the serious data imbalance problem in the modeling process of train vehicle appearance state detection and recognition caused by the limited sample data of abnormal vehicle states, A method based on iterative training combined with feedback correction is used to continuously improve the accuracy of the object detection model, achieving a recognition accuracy of 95.2%.At present, the system has been deployed at a coal transportation station on a certain railway and is running smoothly, improving work efficiency and having high promotion and application value.