OpenEar: an ultra-affordable, high-throughput, and accurate maize ear phenotyping system

Area of Science:

• Agricultural Science
• Computer Science
• Genetics

Background:

• High-throughput crop phenotyping at the single-plant level is crucial for genetic analysis and breeding but remains a significant bottleneck.
• Existing tools for accurate and affordable maize ear phenotyping are limited, hindering progress in the field.

Purpose of the Study:

• To develop OpenEar, an open-source, low-cost phenotyping system for maize ears.
• To create a deep learning-based pipeline for end-to-end extraction of phenotypic data from maize ears.

Main Methods:

• Developed a DIY maize ear imaging platform using 3D-printed parts and readily available electronics for 360° surface scanning.
• Employed Convolutional Neural Network (CNN) models for ear identification and YOLOv11 models for ear segmentation and surface projection image generation.
• Extracted ten key ear and kernel traits using the developed pipeline.

Main Results:

• OpenEar demonstrated high agreement with manual measurements for multiple traits, including ear length (R²=0.972), ear volume (R²=0.976), and kernel number (R²=0.98).
• The system achieved reliable performance across diverse ear and kernel traits, with R² values ranging from 0.515 to 0.98.
• Publicly released manually annotated ear images and videos as a community resource.

Conclusions:

• OpenEar provides a validated, low-cost, and accessible solution for high-throughput maize ear phenotyping.
• The DIY-based approach has the potential to significantly advance crop genetic analysis and breeding applications.