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
- 1 - 10 (10page)
이용수
초록· 키워드
This study presents a deep learning-based recognition system that combines strawberry object detection and pose estimation to deliver real-time recognition information for robotic strawberry harvesting. The system is implemented on a single embedded platform, integrating a Jetson-based computing device with an RGB-D camera. This setup allows for real-time image acquisition, object detection, and keypoint detection in greenhouse environments. Training data were collected from a greenhouse using the RGB-D camera, with three keypoints (fruit center, calyx, and pedicel) annotated in both bounding box and keypoint formats across 557 images. A top-down pose estimation approach utilizing the YOLOv9-pose model was employed. Experimental results demonstrated an object detection performance of 0.94 mAP@0.5, with average keypoint localization errors of 2.54 pixels along the x-axis and 2.88 pixels along the y-axis. When converted to real-world coordinates using depth information, the average spatial errors were 8.57 mm and 2.45 mm. The system achieved an average inference speed of 16.9 fps in the vertical direction, even under varying object density conditions. These results suggest that the proposed system can significantly enhance recognition technologies for automated strawberry harvesting.
#Harvesting robot(수확 로봇)
#Keypoint detection(지점 검출)
#Deep learning(딥러닝)
#Strawberry(딸기)
#Top-down pose estimation(하향식 자세 추정)
상세정보 수정요청해당 페이지 내 제목·저자·목차·페이지정보가 잘못된 경우 알려주세요!
목차
- Abstract
- 1. 서론
- 2. 재료 및 방법
- 3. 결과 및 고찰
- 4. 결론
- References