Detecting Eyes on Potato Tubers with Deep Learning
Author Information
Author(s): Divyanth L. G., Khanal Salik Ram, Paudel Achyut, Mattupalli Chakradhar, Karkee Manoj
Primary Institution: Washington State University
Hypothesis
Can deep-learning-based object detectors effectively identify eyes and stolon scars on potato tubers for robotic sampling?
Conclusion
The study found that YOLOv10m is the most effective model for detecting eyes and stolon scars on potato tubers, balancing accuracy and speed.
Supporting Evidence
- YOLOv10m achieved a mean average precision of 0.911.
- Detection accuracy varied across potato cultivars, with the lowest performance on purple-skinned tubers.
- The study created a robust image dataset of 900 potato tuber images for model training and evaluation.
- Models with fewer parameters showed better inference times without sacrificing accuracy.
Takeaway
This study shows how robots can learn to find the best spots on potatoes to take samples, making it easier to check for diseases.
Methodology
The study evaluated various YOLO-based object detectors on a dataset of potato tuber images to assess their accuracy in detecting eyes and stolon scars.
Limitations
The models struggled with detecting features on purple-skinned potatoes and combined eye and stolon scar detection as a single class.
Participant Demographics
Images from five potato cultivars were used, including russet, red, and purple skinned varieties.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website