High-throughput phenotyping analysis of maize at the seedling stage using end-to-end segmentation network

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• High-throughput phenotyping is crucial for crop monitoring and management.
• Existing image analysis methods often lack accuracy due to reliance on empirical thresholds.

Purpose of the Study:

• To develop an automated phenotype extraction approach for maize seedlings using top-view images.
• To improve the accuracy and efficiency of early-stage crop phenotyping.

Main Methods:

• An end-to-end segmentation network, PlantU-net, was developed for automatic segmentation of maize seedling images.
• The network was trained using a small dataset for efficient learning.
• Morphological and color-related phenotypes were extracted, including shoot coverage, circumscribed radius, aspect ratio, and plant azimuth plane angle.

Main Results:

• PlantU-net accurately segmented maize shoots from top-view images acquired by UAV and tractor-based platforms.
• High average segmentation accuracy (0.96), recall (0.98), and F1 score (0.97) were achieved.
• Extracted phenotypes showed high correlation with manual measurements (R² = 0.96-0.99).

Conclusions:

• The proposed PlantU-net approach offers a practical and expandable solution for high-throughput phenotyping of early-stage crops.
• It requires less training data, enhancing its applicability.
• This method significantly improves the accuracy of phenotype extraction for crop growth monitoring and management.