Inspection of Heat Seal Packing Bag Integrity Using Thermal Images With YOLO Algorithm

This paper presents an inspection of heat seal packing bag integrity using thermal imaging with a deep learning technique. The performance was evaluated by comparing the object detection rates obtained from the YOLOv4-Tiny Algorithm and the Convolutional Neural Network (CNN) technique. Two sets of completely sealed and failure-sealed packaging bags were prepared for the training (100 bags) and testing (100 bags) of the models. Some sample bags containing tomato sauce inserted between the seals represent a failure-sealed condition. The integrity of the heat seal packaging bag was analyzed by examining the differences in color shade patterns of the thermal images. As a result, the YOLOv4-Tiny model achieved a detection accuracy of 98.80%, significantly outperforming the CNN technique, which detected only 78.40%, while using the same dataset; additionally, the time required for detection and training was faster.