CornPheno: Phenotyping corn ear kernels in the wild via point query transformer

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Breeding

Background:

• Manual measurement of corn ear traits is laborious and prone to errors.
• Existing automated phenotyping systems are expensive and require controlled environments.

Purpose of the Study:

• To introduce CornPheno, a smartphone-based system for in-field corn ear phenotyping.
• To enable accurate measurement of kernels per ear, rows per ear, and kernels per row.
• To provide a cost-effective and accessible tool for corn breeders.

Main Methods:

• Utilized a Corn data-trained Point quEry Transformer (CornPET) for kernel prediction.
• Developed a novel point-based row detection method (unicorn) using squeezed clustering and bi-directional point searching.
• Implemented adaptive geometric modeling for robustness against irregular row and kernel patterns.

Main Results:

• CornPheno accurately extracts key corn ear parameters: kernels per ear, rows per ear, and kernels per row.
• The system demonstrates robustness to challenges like partial or curved rows and missing kernels.
• Integration into the OpenPheno WeChat mini-program ensures open access for breeders.

Conclusions:

• CornPheno provides a user-friendly, low-cost, smartphone-based solution for corn ear phenotyping.
• The developed methods (CornPET and unicorn) offer accurate and robust trait extraction.
• This approach facilitates efficient corn breeding by overcoming current phenotyping limitations.