When is a sensor measurement good enough for use in the orchard?
Sensor measurements only become valuable when they help growers and technology partners make better decisions in the orchard. This public summary describes how the Treescout camera system was validated under orchard conditions for blossom, fruit load and vigour. The results show that Treescout already provides useful support for precision thinning and vigour measurements, while measurement timing and fruit load estimation require further improvement.
Innovatiepackage, use case and type test
Open Field Cultivation
Handsfree Orchards
Status: completed
Technical functionality
Broad knowledge question
Making differences between trees visible for more targeted cultivation decisions
Efficient and sustainable fruit cultivation in the Netherlands requires precision sensor methods that can accurately show differences between trees. Within NXTGEN Hightech Agrifood, the study examined how accurately Treescout can measure differences in blossom, fruit load and vigour in orchards.
Approach
Comparison with manual counts tests the value of the measurements
The validation was carried out at three locations: FRC Randwijk and two grower sites. At these locations, Treescout cameras mounted on a tractor collected image data on blossom, fruit load and vigour.
The sensor data were compared with manual counts. At FRC Randwijk, several recording moments on different days and at different times were also compared to better assess the robustness of the measurements.
Goal
Testing accuracy under orchard conditions
The goal of the testing and validation is to assess how accurately Treescout measures blossom, fruit load and vigour under orchard conditions. Use in orchard practice is important because sensor technology only becomes useful when the results provide enough guidance for orchard management decisions.
Result and reflection
Useful measurement accuracy with points for further improvement
In short, Treescout shows useful results for precision thinning and vigour measurements. Further validation is needed for measurement timing, precision root pruning and fruit load estimation.
The results show that the Treescout camera system from Aurea Imaging can correctly classify many trees for precision thinning. For precision root pruning, the vigour map provided useful decision information at the tested location. At the same time, the robustness of measurements at different times of day and the relative estimation of fruit load require further attention.
Successful outcomes
For precision thinning, Treescout can classify around 80% of trees in the correct class, providing a basis for more targeted thinning decisions.
Depending on the action threshold, accuracy ranges from 72 to 90%.
For precision root pruning, the vigour map resulted in a correct decision for 80% of trees at the tested location.
The comparison with manual counts provided valuable validation of the sensor data.
Lessons learned
The robustness of measurement timing during the day requires further attention.
Precision root pruning was only tested at FRC Randwijk, which means broader application requires further investigation.
Fruit load could only be estimated relatively, as more or fewer fruits per tree.
The R² value, ranging from 0.2 to 0.6, shows that the relationship with manual counts is still variable and that further improvement is needed for broader application.