http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
한정수(JEONG-SU HAN),조수현(SOO-HYUN CHO),변대석(DAE-SEOK BYEON),동승훈(SEUNG-HOON TONG) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.11
This study conducted a study to increase the classification accuracy of the alarm type generated in the glass plate division process (hereinafter referred to as the powder plate process). Alarms are classified into a total of 7 types, and there is a problem that misclassification occurs. In particular, intrinsic cracks and chipping are misclassified, causing damage to production. As the crack substrate is classified as chipping and placed in a normal lot, the remaining glass chips of the crack substrate are transferred to the normal substrate. Damage is caused by manual scrapping of the crack substrate and physical and human loss due to normal substrate rework. The reason is that after image shooting, the alarm generation standard was set with quantified specs using Size and Gray Value-Gray (GV). As a result, it was confirmed that classifications of types that were not satisfied with Spec, even if they were of the same type, were misclassified. This paper presents a solution to increase the classification accuracy of types of alarms that cannot be classified as figures after learning by collecting the same types of images through the AI program.