Articles
Advanced methods for yield mapping in tart cherries: tank change tracking and YOLO-DeepSort fruit counting
Article number
1395_38
Pages
289 – 296
Language
English
Abstract
Mapping orchard yield variability is crucial for optimizing production and resource management.
This study presents two methods for yield mapping in tart cherry (Prunus cerasus L.) orchards: tank change tracking and computer vision-based fruit counting.
The first method utilizes proximity sensors, Raspberry Pi, and GPS to estimate yield variability based on tank changes during harvest.
The second method employs computer vision, specifically YOLOv8 and DeepSORT, to count the number of cherries tree‑1. Operators were consistent in tank filling, with an average of 516 kg tank‑1. The average yield for the block was 9.16 t ha‑1. The yield map displayed spatial variations and information collected with the device provided insights into harvest efficiency.
Comparing YOLOv8 versions, YOLOV8n achieved 94% mAP50 mean average precision calculated at an intersection over union (IoU) threshold of 0.50 and 60% mAP50-95 (mean average precision across IoU thresholds ranging from 0.50 to 0.95) after 20 min of training, while YOLOv8x achieved 96% mAP50 and 67% mAP50-95 after 43 min of training.
In terms of inference speed, YOLOv8n required 46.52 ms with average detection confidence of 48%, while YOLOv8x needed 82.88 ms with average detection confidence of 52%. This clearly exemplify the trade-off between speed and precision of the models but highlights the suitability of YOLOv8n for real-time fruit counting.
These methods, utilizing portable computers, offer significant advancements in yield mapping for precision agriculture, providing valuable insights for tart cherry orchards.
This study presents two methods for yield mapping in tart cherry (Prunus cerasus L.) orchards: tank change tracking and computer vision-based fruit counting.
The first method utilizes proximity sensors, Raspberry Pi, and GPS to estimate yield variability based on tank changes during harvest.
The second method employs computer vision, specifically YOLOv8 and DeepSORT, to count the number of cherries tree‑1. Operators were consistent in tank filling, with an average of 516 kg tank‑1. The average yield for the block was 9.16 t ha‑1. The yield map displayed spatial variations and information collected with the device provided insights into harvest efficiency.
Comparing YOLOv8 versions, YOLOV8n achieved 94% mAP50 mean average precision calculated at an intersection over union (IoU) threshold of 0.50 and 60% mAP50-95 (mean average precision across IoU thresholds ranging from 0.50 to 0.95) after 20 min of training, while YOLOv8x achieved 96% mAP50 and 67% mAP50-95 after 43 min of training.
In terms of inference speed, YOLOv8n required 46.52 ms with average detection confidence of 48%, while YOLOv8x needed 82.88 ms with average detection confidence of 52%. This clearly exemplify the trade-off between speed and precision of the models but highlights the suitability of YOLOv8n for real-time fruit counting.
These methods, utilizing portable computers, offer significant advancements in yield mapping for precision agriculture, providing valuable insights for tart cherry orchards.
Authors
A. Safre, A. Torres-Rua, B. Black, B. Schaffer
Keywords
fruit recognition, object detection, Prunus cerasus L., Raspberry Pi, yield variability
Groups involved
- Division Temperate Tree Nuts
- Division Vine and Berry Fruits
- Division Temperate Tree Fruits
- Division Precision Horticulture and Engineering
- Division Plant-Environment Interactions in Field Systems
- Division Tropical and Subtropical Fruit and Nuts
- Working Group Precision Management of Orchards and Vineyards
Online Articles (59)