Publication

Good quality packaging prevents contamination, secures preservation, and increases the ease of transportation in food and medical industries. One particular weakness of the package lies in the seal region where contents can be unintentionally incorporated, which disrupts the sealing process and compromises the structure and durability of the seal. To validate the seal quality effectively at high speed, a non-destructive high-resolution inspection approach combining enhanced sensors and reconstruction techniques is required. As the seal is flat and defects are minuscule, sensors have to be placed along the contour of the seal to achieve sufficient sensitivity. However, such conformal sensor placement poses new challenges to the ill-posed traditional tomography reconstruction. To overcome the limitation of sensing angle projections, imbalance in pixel representation and physical measurements, and asymmetric geometry of the sensed region, we propose a high-speed supervised autoencoder reconstruction approach. In this paper, our approach achieves high reconstruction image quality of irregular seal regions despite conformal sensor placement. While overcoming the limitations faced in traditional tomography, our model can be seamlessly integrated into the production line for real-time defect detection without affecting production speed and effectively minimizing manufacturing wastage and downtime.

Researcher/Author: Jieming Pan, Zaifeng Yang, Stephanie Hui Kit Yap, Xiangyu Zhang, Zefeng Xu, Yida Li, Yuxuan Luo, Evgeny Zamburg, En-Xiao Liu, Chen-Khong Tham, Aaron Voon-Yew Thean

Food Packaging and Shelf Life, Volume 33, 2022, 100919,