Character image blender (CIB) data augmentation model and autonomous container character recognition system for smart harbor

Abstract

Character recognition on container boxes has been limited to the side view of containers. Moreover, existing systems suffer from mislocated containers in the yards or ships in ports. Therefore, it is crucial to identify characters before cranes lift containers for more advanced automation. Based on the above needs, Automated Container Identification (ACID) as a container character reading system, is proposed using a deep-learning model for actual implementation on cranes for more autonomous and efficient operation in ports. It robustly recognizes characters from above the container in real-time. ACID consists mainly of three components. First, a character image blender (CIB) is developed to extract data features using data enrichment technology so as to increase the data volume and recognition accuracy with the help of generative adversarial neural networks (GANs). Second, a character-recognition (CR) model is designed to read top-surface images delivered by means of real-time vision streaming. Third, metrics for the generated character images and the predicted character sequences by ACID, are devised to evaluate the proposed method. ACID achieves a recognition rate of 97.31%, with proper consideration of the positions of characters and 15 fps on average for real-time application.