P‐135: The AI based auto masking technique for detection of small stage vacuum hole †

Abstract

In order to improve productivity in the display lamination process, it was necessary to find the vacuum holes on the stage and perform masking treatment. Although the vacuum holes were detected by employing a conventional image processing technique, there was a problem that it was impossible to detect all of them. Therefore, we required the AI object detection technology that can solve the problem. However, the hole has a size of 11 × 11 px and corresponds to the tiny size which is the most difficult for small object detection. For the development of AI auto masking technology, YOLOv5 was selected as the AI object detector because it was light, fast and had good performance, and SAHI technology, which had excellent performance in detecting small objects and excellent compatibility with the various detectors, was selected. The hyper‐parameters tuning and various optimization approaches were performed on AI techniques and the AI auto masking technology was developed. As a result, the time to detect vacuum hole was reduced to 20% compared to the conventional image processing technique. In the SAHI technique, the direct proportion between the slice width/height size and bounding box size of the detected object was found, and the application of the image based accurate object size was also discussed.