Phenotyping of Drought-Stressed Poplar Saplings Using Exemplar-Based Data Generation and Leaf-Level Structural...

Area of Science:

• Plant Science
• Computer Vision
• Machine Learning

Background:

• Drought stress significantly threatens poplar growth and yield.
• High-throughput plant phenotyping offers rapid, non-destructive analysis of plant health.
• Accurate phenotyping is crucial for identifying drought-resilient crops.

Purpose of the Study:

• To develop and evaluate computer vision and deep learning methods for drought-stressed poplar sapling phenotyping.
• To enable accurate leaf posture calculation and drought stress identification.
• To assess the performance of multitask learning models for simultaneous variety and stress level determination.

Main Methods:

• Instance segmentation was employed to extract leaf, petiole, and midvein regions from color images.
• A dataset augmentation technique was developed to minimize manual annotation efforts.
• Horizontal angles of petioles and midveins were calculated for leaf posture digitization.
• Multitask learning models, including MobileNet, were utilized for stress level and variety classification.

Main Results:

• Leaf posture was digitized by calculating horizontal angles of petioles (MAE 10.7°) and midveins (MAE 8.2°).
• Drought stress was found to increase the horizontal angle of poplar leaves.
• The multitask MobileNet model achieved high accuracy (99% for variety, 76% for stress level), outperforming single-task models.

Conclusions:

• The proposed computer vision and deep learning approach effectively enables drought stress phenotyping in poplar saplings.
• This methodology can be applied for screening drought-resistant poplar varieties.
• The findings support precise irrigation decision-making for improved crop management.