Image-Based High-Throughput Detection and Phenotype Evaluation Method for Multiple Lettuce Varieties.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• Assessing lettuce yield and quality traditionally relies on manual phenotyping, which is labor-intensive and time-consuming for numerous varieties.
• Automated, high-throughput phenotyping is crucial for accelerating crop improvement and understanding plant growth dynamics.

Purpose of the Study:

• To develop and validate a novel, non-invasive, high-throughput phenotyping method for lettuce using convolutional neural networks (CNNs).
• To enable accurate assessment of growth rate and quality traits in multiple lettuce varieties through image analysis.

Main Methods:

• A "Sensor-to-Plant" platform captured top-view images of over 2000 lettuce plants across 500 varieties at eight growth stages.
• A multistage CNN model integrated object detection (99.82% accuracy) for pot identification and semantic segmentation (97.65% F1 score) for plant delineation.
• A phenotyping pipeline extracted 15 static traits (geometry, color) and calculated dynamic traits (growth, accumulation rates).

Main Results:

• The CNN-based method accurately identified individual lettuce plants and their pots from image data.
• Extracted static and dynamic traits showed strong correlation with digital biomass and quality indicators.
• Accumulation rates of static traits provided a more accurate reflection of lettuce growth status compared to static traits alone.

Conclusions:

• The proposed image-based phenotyping method offers a non-invasive, high-throughput solution for evaluating lettuce growth and quality.
• This CNN-driven approach can be extended to other crops like maize, wheat, and soybean for efficient phenotype evaluation.