인문학
사회과학
자연과학
공학
의약학
농수해양학
예술체육학
복합학
DBpia AI
AI 에이전트
AI 아이디어
AI 리더
AI 뷰어
Citeasy
크롬 서지관리 다운로드
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.
학술저널
Full-text
오류 신고하기해당 페이지 내 제목·저자·목차·페이지정보가 잘못된 경우 알려주세요!
초록·키워드
The construction industry faces labor shortages and safety concerns, necessitating the automation of heavy equipment. However, unlike structured environments like mines, construction sites are dynamic and complex, posing significant challenges for vision-based autonomous systems. A major limitation is the lack of high-quality training data, especially for critical work target objects such as steel pipe piles and rebar, which are vital for equipment operation yet difficult to capture.
This study introduces a synthetic data generation methodology utilizing the FLUX.1-schnell text-to-image model, with a focus on steel pipe piles as the target object for pile drivers. Through prompt engineering across ten scenario categories, we generated 1,000 synthetic images, from which 270 high-quality images were selected using CLIP Score-based filtering. These synthetic images were then combined with 55 real-world images to train a YOLOv8n object detection model. Independent testing on 30 unseen construction site images revealed a significant performance disparity: the synthetic-only model achieved near-perfect validation performance (mAP50 0.995) but failed completely in real-world testing (0% detection rate). In contrast, the hybrid model, despite having lower validation performance (mAP50 0.912), achieved an 83.3% detection rate and 62.3% average confidence in real-world scenarios. This study demonstrates that generative AI-based synthetic data can effectively address the data scarcity in construction domains. Moreover, it highlights the importance of strategically combining large-scale synthetic data with a small amount of real data (16.9%) to bridge the domain gap and enhance real-world applicability.
This study introduces a synthetic data generation methodology utilizing the FLUX.1-schnell text-to-image model, with a focus on steel pipe piles as the target object for pile drivers. Through prompt engineering across ten scenario categories, we generated 1,000 synthetic images, from which 270 high-quality images were selected using CLIP Score-based filtering. These synthetic images were then combined with 55 real-world images to train a YOLOv8n object detection model. Independent testing on 30 unseen construction site images revealed a significant performance disparity: the synthetic-only model achieved near-perfect validation performance (mAP50 0.995) but failed completely in real-world testing (0% detection rate). In contrast, the hybrid model, despite having lower validation performance (mAP50 0.912), achieved an 83.3% detection rate and 62.3% average confidence in real-world scenarios. This study demonstrates that generative AI-based synthetic data can effectively address the data scarcity in construction domains. Moreover, it highlights the importance of strategically combining large-scale synthetic data with a small amount of real data (16.9%) to bridge the domain gap and enhance real-world applicability.
인공지능 문자 인식 모델을 통해 추출된 텍스트로, 일부 오타나 오류가 포함될 수 있으나 지속적으로 개선 중입니다.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
최근 본 자료 전체보기
UCI(KEPA) : I410-151-26-02-094786673
