Key points are not available for this paper at this time.
Shipping containers play a pivotal role in both local and international transportation. This paper presents a deep learning-based solution for localizing and recognizing general-purpose shipping container codes. The container code localization model, leveraging ResNet and U-Net, processes a rear-view container image to identify three bounding boxes for code components (owner code & category identifier, serial number, and check digit). These bounding boxes guide image cropping, and the segmented images are then input into the recognition model based on CNN and RNN, converting the image text into a machine-readable format. Experimental results showcase the proposed system's impressive accuracy of 95%, surpassing state-of-the-art models such as EAST, PSENET, GCRNN, and MaskTextSpotter. With potential applications in port terminals, where precise and efficient code recognition is crucial, the proposed system offers operational streamlining, workflow efficiency enhancement, and improved overall productivity in container handling processes.
Hlabisa et al. (Thu,) studied this question.