Beef Tomatoes Reach 98.8% Fruit Counting Accuracy

FRAVEBOT’s AI-driven fruit counting has achieved a new milestone. During a four-month trial with a Dutch tomato producer, the Scout reached 98.8% accuracy in detecting and counting beef tomatoes. This success proves the robustness of our training pipeline, enabling new crop segments to be calibrated and deployed within just three weeks.

We are proud to announce a major milestone in our fruit counting and ripeness assessment technology. During a four-month trial with a leading Dutch tomato producer, our AI model achieved 98.8% accuracy in counting beef tomatoes.

But the achievement goes beyond one crop. We have developed a robust training pipeline that allows us to introduce a new fruit segment and reach operational accuracy within just three weeks after deployment.

Fruit Validation Methodology

How We Got There

In the initial phase, our team fine-tuned the fruit counting model to exceed 95% accuracy and validated it over a two-week observation period. In the second stage, we adapted the ripeness scale to the grower’s specific requirements — aligning the model with internal grading practices instead of the standard seven-stage textbook scale (breaker, turning, light red, red, etc.).

Adapting to Real-World Conditions

Training involves both daytime and nighttime scanning to capture diverse lighting conditions. For greenhouses equipped with LED lighting, our algorithms also compensate for color spectrum shifts, ensuring that ripeness detection remains consistent and accurate under any conditions.

What’s Next

We are now expanding the FRAVEBOT Scout’s capabilities into other crops for production monitoring.

Read about our recent developments in pepper production to see how the same data-driven approach is transforming multi-crop greenhouse management.