Traditionally, fruit counting in the field has been carried out using manual tools, which involves a high labour requirement, high management times and uncertainty in the final estimates due to the need for sampling. However, this is a necessary activity to optimise the management of an agricultural plantation, which is why automated systems are being developed that can carry out this task. These commercial systems include high-cost technologies, so the implementation rates in the field are not very high.
In this context, researchers at the Universitat de Lleida (UdL) are developing a system based on artificial intelligence. It is designed for the detection and calibration of fruit in the field that will provide a more accessible alternative. The system is based on the combination of RGB-D optical sensors together with an innovative multi-tasking deep neural network (DNN). This artificial intelligence system has been developed and trained for the detection and sizing of fruits.
The technology is in the development phase and has been tested experimentally in apple fields, obtaining results of 90% accuracy in fruit counting and 5 mm error in size estimation.
This technology has some competitive advantages compared to similar technologies:
- Cost-effective technology that allows simultaneous counting and sizing of fruit.
- Reduction of the error rate in sizing compared to manual measurements, as well as a reduction in execution time and human resources involved.
- The system allows yield predictions to be made, resulting in more efficient harvest and storage management.
- The hardware system is small in size and adaptable to commonly used agricultural machinery, allowing it to work in parallel with other agricultural operations.
The represented institution is looking for a collaboration that will lead to a further development of the technology considering the needs of the sector.
Institution: Universitat de Lleida
Contact: Noelia Mas / firstname.lastname@example.org