ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Smart Greenhouse Rose Monitoring: Bud Stage Recognition and Leaf Area Estimation

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600611.(doi:10.13031/aim.202600611)
Authors:   Han-Sheng Chang, Yu-Shan Wang, Wei-Ling Chen, Hsing-Ying Chung, Shih-Fang Chen
Keywords:   Precision agriculture, Rose growth monitoring, Rail-mounted imaging system, 3D reconstruction, Canopy structure analysis

Abstract.

Roses (Rosa hybrida) are among the most economically valuable floriculture crops, with production highly sensitive to complex interactions of environmental factors including temperature, humidity, light intensity, CO₂ concentration, and crop management practices. Optimizing greenhouse conditions remains challenging because growers are typically limited to assessing total yield per production zone, unable to identify optimal parameters for specific cultivar-environment combinations. Bud stage recognition enables tracking of reproductive development and yield formation rates under varying conditions, while leaf area quantification provides assessment of vegetative growth status, photosynthetic capacity, and resource allocation patterns. This study developed an automated monitoring system using a ceiling-mounted rail-based imaging platform integrated with deep learning and 3D reconstruction to achieve spatially-resolved bud stage recognition and leaf area quantification correlated with localized environmental conditions. For bud stage recognition, a YOLOv11 model trained on approximately 900 manually annotated images of 'Laventine' and 'Beehive' roses simultaneously localize buds and classifies four developmental stages: visible bud formation, bud enlargement, petal exposure, and harvest maturity. The models achieved mAP50 of 0.805 and 0.870, respectively. Primary challenges include detecting early-stage buds and transition periods between later stages, expected to improve with additional training data. For leaf area quantification, single-view videos from a top-down perspective were processed using a sparse-view 3D reconstruction framework. Leaf area quantification assesses management practices balancing photosynthetic capacity versus nutrient competition, and investigates vegetative-reproductive relationships for optimizing source-sink balance. To investigate the temporal relationship between canopy structure and reproductive development, cross-correlation function (CCF) analysis was conducted between estimated total leaf area and the percentage distributions of different bud developmental stages. The results revealed stage-specific lagged relationships, with the B1 stage showing the strongest positive correlation at a lag of -6 days (r = 0.815, p = 0.093), suggesting that increases in canopy leaf area preceded higher proportions of early-stage buds. This integrated system provides spatially-explicit data linking bud stage progression and leaf area to localized environmental conditions, establishing a foundation for further plant-environment interaction analysis and advancing precision floriculture through data-driven optimization.

(Download PDF)    (Export to EndNotes)