Study finds strawberry colour can mislead harvesting robots
Researchers at the University of East Anglia (UEA), working with agri-tech firm Antobot, have found that the colour of strawberries captured by cameras may not always provide an accurate indication of ripeness, raising questions about the reliability of vision-based agricultural robotics.
As automation becomes more common in crop monitoring and harvesting, cameras are increasingly used to assess fruit ripeness, detect disease, estimate yields, and support quality control. For strawberries, ripeness is largely judged by colour.
Prof Graham Finlayson, from UEA’s School of Computing Sciences, said: “In the race to modernise agriculture, farmers are increasingly relying on technology to monitor plant health or predict crop yield.
“Cameras have taken on jobs once done by the human eye – for example, they may be monitoring crop health, detecting disease, estimating yields, deciding when fruit is ready to harvest or determining what ends up on supermarket shelves.”
To investigate the accuracy of camera-based assessments, researchers photographed strawberries under field conditions while the fruit remained on the plant. They then compared the images with berries selected for harvest by experienced pickers.
Postgraduate researcher James Bennett said: “When we processed images using standard camera settings, the strawberries showed noticeable differences in colour – even when they were equally ripe.
“But after applying a simple calibration technique using a colour checker placed in each image, much of that variation disappeared.”
According to the researchers, uncorrected images showed ripe strawberries ranging from deep magenta to bright orange-red depending on lighting conditions, while calibrated images produced more consistent colour measurements.
The calibration method reduced variation in strawberry colour measurements by 48 per cent, indicating that much of the inconsistency originated from the imaging process rather than the fruit itself.
The findings suggest that some colour differences detected by automated systems may be caused by image capture and processing conditions rather than actual variation in ripeness.
Prof Finlayson said: “As automation becomes more widespread, ensuring the fidelity of visual data will be crucial for reliable decision-making – from yield predictions to quality control.”
Marc Jones of Antobot added: “Autonomous robots are only as good as the information they receive. Improving the consistency and reliability of crop data is fundamental to making better decisions in the field, whether that’s monitoring crop health, identifying harvest readiness or improving yields.
“This research is an important step towards building more intelligent, more trusted and ultimately more productive autonomous farming systems.”
For more information:
Lisa Horton
University of East Anglia
Email: l.horton@uea.ac.uk
