닫기
Loading..

전자정보연구정보센터 ICT 융합 전문연구정보의 집대성

학술대회 프로시딩

홈 홈 > 연구문헌 > 학술대회 프로시딩 > 한국정보과학회 학술대회 > KSC 2018

KSC 2018

Current Result Document : 287 / 803 이전건 이전건   다음건 다음건

한글제목(Korean Title) 작은 데이터셋에서의 복잡한 금속 부품 결함 검출
영문제목(English Title) Defect detection of complicated metal parts with a small dataset
저자(Author) Jahongir Yunusov   Bharat Subedi   Abdulaziz Gaybula   김태형   Tae-Hyong Kim  
원문수록처(Citation) VOL 45 NO. 02 PP. 0848 ~ 0850 (2018. 12)
한글내용
(Korean Abstract)
영문내용
(English Abstract)
Defect detection and localization is a crucial part in the product inspection stage. Although many research studies and advances in the object detection area with deep learning technology, there are still not a few challenges in defect identification, especially in case that products are in complicated shapes whose boundaries may be misidentified as scratches in learning process. A limited amount of dataset is another obstacle, which easily leads to overfitting in deep learning networks. This paper tries to solve this problem by rather sophisticated preprocessing of dataset. With preprocessed images, our capsule network could learn defects well without overfitting. The Grad-CAM technology verified that the network localized defective areas correctly.
키워드(Keyword)   
원문 PDF 다운로드