A Study of Introducing Real-time Object Detection Algorithm into Screen Printing Thermal Transfer Printed Label Defect Detection

This research focuses on the necessity and feasibility of improving traditional optical inspection systems and using advanced object detection algorithms to detect industrial defects. Through actual inspection test results, it can successfully detect the content of PET film thermal transfer labels. Defect characteristics. The main architecture of this research includes the YOLO-V3 algorithm and the EfficientDet algorithm. Clear images obtained by planning and analyzing hardware equipment for defect accuracy and characteristics are used as functional verification. This research has achieved good research results, with the calculation accuracy reaching mAP 99% and meeting the speed requirement of real-time defect detection of 30fps (above). The introduction of the inspection system can increase the production capacity of the quality inspection department. In the past, each inspector produced an average of 3,500 labels. With the addition of the AI inspection system, the number can be increased to 5,500 labels. In addition to saving human resources, increasing production capacity can also share fatigue among inspection personnel.