VNExpirationDate: A Novel Expiration Date Dataset and a Robust Framework for Expiration Date Recognition on Complex Packaging in Vietnam

The Food and Agriculture Organization of the United Nations reports that approximately 1.3 billion tons of food produced for human consumption is wasted annually [1], with a significant portion due to unawareness of expiration dates at the consumer level, where food is often discarded due to spoilage or overlooked expiration dates. Although some computer vision solutions have been developed to address this issue, they face limitations stemming from the complexity of packaging designs and the diversity of date formats, particularly in regions like Vietnam. To overcome these challenges, this study introduces the Vietnamese Varifocal and Autonomous (VVA) model, a new approach tailored for expiration date detection and recognition on the distinct packaging labels in Vietnam's market. VVA integrates an IoU-aware dense object detector for precise date localization with a bidirectional, iterative model for robust date recognition. Additionally, we present VNExpirationDate, a novel dataset that captures the unique blend of local and global products in Vietnam, supporting 13 distinct date formats. Experimental results demonstrate that our approach achieves a high accuracy of 96.7% in expiration date recognition.