인문학
사회과학
자연과학
공학
의약학
농수해양학
예술체육학
복합학
지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학술저널
- 저자정보
- 발행연도
- 2026.6
- 수록면
- 239 - 244 (6page)
이용수
초록· 키워드
Automation of complex post-processing in large-scale manufacturing can substantially reduce human risk while improving operational efficiency. Real-time, fine-grained recognition and segmentation of objects and their detachable components are essential for such automation, yet task- specific dataset construction remains challenging due to the diversity and scale of industrial objects. To overcome these challenges, we propose a simulation-to-reality pipeline that leverages high-fidelity synthetic data generated in Unreal Engine 5. Our method produces photorealistic images and auto- matically generates pixel-accurate annotations at scale, eliminating manual labeling. The resulting datasets are used to train YOLOv8-based detection and segmentation models, which are evaluated both in zero-shot transfer settings and through few-shot fine-tuning with limited real samples. Model performance is assessed on real-world imagery using standard metrics with attention to real-time inference requirements crucial for industrial deployment. Our results demonstrate that synthetic data can effectively bridge the gap between virtual and real domains, providing a scalable framework for vision-based automation in complex manufacturing environments.
상세정보 수정요청해당 페이지 내 제목·저자·목차·페이지정보가 잘못된 경우 알려주세요!
목차
- Abstract
- 1. 서론
- 2. 관련 연구
- 3. 본론
- 4. 실험
- 5. 결론
- References