A Hybrid Model for Container-code Detection

In this paper, we propose a container code detection algorithm that combines PSENet and CRNN in a manner that ensures that the obtained images are greatly affected by the other containers as well as text information in the picture. The proposed algorithm divided into three parts: object detection module, text detection module and text recognition module. At the initial step, the object detection module is used to calculate the position of those areas in container code that needs to be predicted, we then use the text detection module based on pixel segmentation, and finally obtain the container code through the end-to-end text recognition module. This algorithm is able to detect different container code for a vertical and horizontal scaling. We show that, in a complex multi-container scenario, the detection performance is good, and highly stable when trained with multi-angle container, achieving relative improvement of up to 95%.