Open Collections

Bahrami, Zhila

University of British Columbia

Shipping containers have revolutionized global trade and become vital to supply chain infrastructure. However, shipping containers must be regularly inspected to ensure contents are appropriately protected. Shipping container inspection refers to monitoring containers’ security and safety conditions during transit; containers have not been tampered with and are maintained safely during their life cycles. Currently, the inspection of containers is performed manually by human observation, which is error-prone and time-consuming. This research aims to provide automated shipping container inspection solutions that benefit the global transportation industry. First, the security of shipping containers is inspected. A deep learning-based architecture is developed to examine security seals on the container’s rear for automated security inspection of shipping containers. It consists of two key modules: a multi-scale multi-depth image pyramid network to extract feature representations and an attention-based memory bank to extract long-range spatial dependencies. These two modules provide the terminals with a highly accurate and efficient security inspection solution. Then, the safety condition of containers is thoroughly inspected to maintain containers in good condition. Finally, a supervised deep learning-based architecture is proposed to detect and characterize defects on the surface of shipping containers. The proposed framework employs a multi-scale multi-depth image pyramid network in conjunction with two attention-based memory banks to achieve high performance on defect detection. Additionally, the framework introduces a novel optical flow-based image stitching for defect characterization to estimate the percentage of defects present on the surface of a container. Last but not least, an unsupervised deep learning-based architecture is developed to inspect the safety of shipping containers. A considerable number of data and their ground truth are demanded to be prepared by human annotators, which is costly and time-consuming. Our framework adopts defect generation and feature extraction using multi-scale multi-depth networks and Siamese networks to detect abnormal regions on the surface of containers. The proposed framework is well-generalized and works well at terminals with various conditions. The outcomes of this research will benefit the global transportation industry since they will assist in the standardization and reinforcement of container management and logistics at terminals.

Text

2023-01-05

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/83593

Electrical Engineering

University of British Columbia

UBCO

http://creativecommons.org/licenses/by-nc-nd/4.0/