Development of double-camera AI system for efficient monitoring of paprika fruits

In high-tech greenhouses, yield prediction contributes to environmental control and efficient worker management, price negotiation, and shipment planning.
To predict crop yields, some large-scale greenhouse growers manually count the number of fruits on some plants weekly.
In our previous study, we proposed a fruit monitoring system for paprika plants using deep learning-based object detection (Mask R-CNN). The scanning device, attached to a pipe rail trolley, moves at a stable speed along a pipe rail between the plant rows, scanning the crop canopy in a plant row, creating a panoramic image, and automatically counting the number of fruits.
Previously, the system scanned one plant row while moving forward and the other row while moving backward.
In this study, we updated our device by mounting two cameras facing opposite directions, maintaining a size of 1.3 m (H) × 0.4 m (W) × 0.25 m (D), which is close to the size of the previous device.
This modification enabled us to scan crop canopies in both plant rows simultaneously, saving measurement time.
We also automated the data transfer and analysis process so the grower can receive the monitored results promptly.
We tested our system in a large-scale commercial greenhouse (with a total cultivation area of approximately 2.4 ha) by monitoring the number of paprika fruits and examining the relationship between the number of detected fruits, yield, and harvesting time.
Our results showed that the number of detected paprika fruits correlated with the yield and harvesting time of the following week.

International Symposium on New Technologies for Sustainable Greenhouse Systems: GreenSys2023

K. Shimomoto, M. Shimazu, T. Matsuo, S. Kato, H. Naito, T. Fukatsu

greenhouse horticulture, object detection, number of fruits, simultaneous measurement, yield prediction

https://doi.org/10.17660/ActaHortic.2025.1426.49