Best-View Pedicel Localization with YOLO-DSC for Calyx-Preserving Robotic Harvesting of Cherry Tomatoes
Verianti Liana, Hao Cheng Zuo, Ping-Lang Yen
AI summary
Problem
Robotic cherry tomato harvesting struggles with dense foliage, asynchronous ripening, and strict calyx-preservation requirements that obscure the thin pedicel. Existing vision systems fail under dynamic viewpoints, calibration drift, and false positives from visually similar distractors.
Approach
The system actively searches for the optimal camera viewpoint to maximize pedicel visibility, uses a YOLO-DSC detector trained with distractor data to suppress false positives, and confirms the cutting point with a laser module.
Key results
- YOLO-DSC detector with null-data training boosts pedicel localization precision to 0.97
- Best-view searching strategy maximizes visible pedicel length and separates it from the calyx
- Laser confirmation provides robust cutting point positioning under occlusion
- Real-world deployment achieves 73.3% harvesting success, a 26.3% improvement over baseline
Why it matters
Enables reliable, market-compliant robotic harvesting of cherry tomatoes in unstructured greenhouses, advancing automation for high-value specialty crops.
Abstract
Robotic harvesting of cherry tomatoes remains challenging due to dense foliage, asynchronous ripening, and the strict market requirement for calyx-preserving cuts. The calyx frequently occludes the pedicel, making precise localization indispensable. In 640 × 480 images, pedicels span only 7–32 pixels, where even minor errors can lead to miscutting the calyx. To address this challenge, we apply YOLO-DSC to localize pedicels across dynamic frames as the arm-mounted camera moves during the best-view search. This strategy maximizes the visible pedicel length, exposing it perpendicularly to the camera and ensuring clear separation from the calyx, while null-data suppresses false positives from distractors such as leaves, stems, and calyces. In 15 autonomous trials along a 28 m greenhouse row, YOLO-DSC achieved the lowest pedicel localization errors, outperforming the YOLO baseline model (p < 0.05). This improvement directly translated into higher harvesting success, increasing from 47% with YOLO (including null data training) to 73.3% with YOLO-DSC. These results demonstrate that integrating YOLO-DSC with best-view searching enhances recall and stability under dynamic viewpoints, enabling more reliable calyx-preserving harvesting in real greenhouse conditions.