[SCIE] Improved U-Net++ with Patch Split for Micro-defect Inspection in Silk Screen Printing
- 윤병관
- 조회수1484
- 2022-03-23
-Title: Improved U-Net++ with Patch Split for Micro-defect Inspection in Silk Screen Printing
-Journal/Conference: MDPI Applied Sciences Journal (Special Issue: Applications of Deep Learning and Artificial Intelligence Methods), 12(9), 4679, 6 May 2022
-Authors: Byungguan Yoon, Homin Lee and Jongpil Jeong *
-DOI: https://doi.org/10.3390/app12094679 / https://www.mdpi.com/2076-3417/12/9/4679
-Journal/Conference Link: https://www.mdpi.com/journal/applsci/sections/computing_artificial_intelligence
Abstract:
Due to the trend of multi-variety production, silk screen prints are also changing the type of product at short intervals. Since the type and location of defects that usually occur change, it is difficult for the operator to conduct a quality inspection for fine defects. In this paper, we propose an improved U-Net++ based on patch split for the purpose of automated quality inspection for fine defects. Instead of using Input of the original image level, it is a method of processing defects in the image well by using Input of the Patch level. In the case of using the existing Original Input, learning was not performed smoothly. On the other hand, the method we proposed was stably learned and the Ice Score was the highest at 0.728. Products with fine defects compared to product size, such as silk screen prints, will also be able to be detected. In addition, it was shown that quality inspection using Patch Split Method-based AI is possible even in situations where there is little defective data.