• Depth informed trunk detection performed reliably under field conditions. • Trunk diameter estimates aligned closely with ground truth. • Transmission-based digital odometer provided reliable along-row positioning. • Full-block mapping demonstrated large-scale applicability. Trunk diameter is an important structural trait used to assess tree size, vigor, and long-term growth in orchard systems, but it remains difficult to measure efficiently at orchard scale. This study developed and validated a low-cost mobile imaging system for automated trunk diameter estimation and spatial mapping in tart cherry (Prunus cerasus) orchards. The system integrated RGB-D cameras, deep learning-based trunk detection and segmentation, three-dimensional depth reconstruction, and driveline-based digital odometry for spatial positioning under canopy conditions where GNSS performance was unreliable. Trunks were detected using a YOLO-based model, segmented using a Segment Anything Model (SAM)-based approach, and reconstructed in 3D from depth data to estimate real-world trunk diameter at 40 cm above the trunk base. The system was evaluated in a 15-year-old, 1 ha experimental orchard in Utah, USA, and then applied in a 14-year-old, 9.5 ha commercial orchard. Under unobstructed viewing conditions, trunk diameter estimates showed strong agreement with manual measurements, with a mean absolute error (MAE) of 0.95 cm, a root mean square error (RMSE) of 1.16 cm, and R 2 of 0.79. Across all 462 trees, automated per-tree averaging produced an MAE of 1.40 cm. In the commercial orchard, mapped trunk diameter patterns aligned with UAV-derived canopy height, reflecting underlying zones of tree vigor. These results show that mobile RGB-D imaging combined with driveline-based odometry can provide practical, cost-effective orchard-scale trunk diameter mapping under commercial field conditions.
Wedegaertner et al. (Wed,) studied this question.