This master’s thesis, authored by Sijun Yu at the Cyprus University of Technology, investigates the automated detection of broken seals on containers using video recordings and deep learning techniques. The work is situated within the broader context of supply chain security, where ensuring the integrity of container seals is critical for preventing tampering, theft, and unauthorized access. The thesis leverages the YOLO (You Only Look Once) model, a state-of-the-art, real-time object detection framework, to address the challenges associated with identifying compromised seals in diverse and potentially complex visual environments.
Application of YOLO to Seal Detection: The thesis adapts the YOLO model for the specific task of detecting broken seals on shipping containers from video footage. This involves customizing the model architecture and training process to recognize subtle visual cues indicative of seal breakage, which can be challenging due to varying lighting, angles, and occlusions.
Dataset Creation and Annotation: A significant contribution is the assembly and annotation of a dataset comprising video frames or images of container seals in both intact and broken states. This dataset forms the foundation for training and evaluating the detection model, ensuring that it can generalize to real-world scenarios.
Performance Evaluation: The thesis rigorously evaluates the adapted YOLO model’s performance, likely using metrics such as precision, recall, and mean Average Precision (mAP). The results demonstrate the model’s effectiveness in accurately and efficiently identifying broken seals, highlighting its potential for deployment in operational settings.
Automation and Real-Time Analysis: By utilizing YOLO’s real-time detection capabilities, the proposed system enables automated, continuous monitoring of container seals from video streams, reducing the need for manual inspection and increasing the reliability of security checks.
The research presented in this thesis has significant implications for logistics, transportation, and supply chain security. Automated detection of broken seals enhances the ability of organizations to quickly identify and respond to security breaches, minimizing losses and maintaining regulatory compliance. The use of deep learning and video analysis represents a modern, scalable approach that can be integrated into existing surveillance infrastructures.
Furthermore, the thesis contributes to the growing body of work applying computer vision and artificial intelligence to industrial and security applications. By demonstrating the adaptability of the YOLO model to a specialized detection task, the research opens avenues for further enhancements, such as incorporating temporal information from video sequences, improving robustness to environmental variability, and expanding to other forms of tamper detection. Overall, the thesis exemplifies the practical benefits of AI-driven automation in critical security domains.