Researchers have unveiled a potato scanner that uses artificial intelligence to improve the quality-control process.
The team from Lincoln University used off-the-shelf hardware and new software to build a low-cost system that identifies defects, diseases and blemishes in real time, and can be reprogrammed for different potato types and operating conditions.
The aim of the project, which was carried out with the aid of industry body the Potato Council, was to create a system that could augment the work of quality-control (QC) staff by improving the consistency, speed and accuracy of defect detection.
‘Most potatoes are still sorted by hand,’ Dr Tom Duckett, director of Lincoln’s Centre for Vision and Robotics Research, told The Engineer.
‘Problems with manual sorting include the subjectivity, fatigue and high cost of human inspectors, while currently deployed artificial-vision systems require manual calibration and have limited accuracy.’
The system comprises a low-cost vision sensor and standard desktop computer. This uses software that takes input from human experts to learn how to identify differences in colour and texture between blemished and unblemished skin in a specific sample.
To enable the software to deal with the large amounts of natural variation in the produce, the researchers created a machine-learning algorithm to automatically select good image features.
The other major challenge was enabling the system to work fast enough to analyse the potato in real time, as the original software took several hours for each image.
Duckett’s team is now seeking funding to commercialise the technology. ‘If the current funding bid is successful, we should see the first commercial systems being ready for market within the next three years,’ he said.
The team also hopes to develop the technology into a system that checks every potato on the production line, at lower cost and with higher accuracy than current equipment.