Stem water potential estimation from images using a field noise-robust deep regression-based approach in peach trees

Area of Science:

• Agricultural Science
• Plant Physiology
• Computer Vision

Background:

• Field-grown peach trees present challenges for water deficit stress detection due to their size and complex structures.
• Accurate monitoring of tree water status is crucial for optimizing irrigation and managing crop yield.

Purpose of the Study:

• To develop and validate a deep-learning-based method for precise detection of water deficit stress in large, field-grown peach trees.
• To establish a robust system for diagnosing water status by analyzing image data.

Main Methods:

• Acquired extensive image datasets from field-grown peach trees under varying water status conditions induced by partial secession treatment.
• Utilized a deep-learning regression model to predict stem water potential (Ψstem) from thousands of video frames.
• Implemented an averaging method on predicted Ψstem values from video frames to represent the water status of individual branches.

Main Results:

• The deep-learning model achieved a high coefficient of determination (0.927) between measured and predicted stem water potential (Ψstem) after averaging frame predictions.
• The averaging technique effectively reduced noise and provided a representative water status value for each tree.
• The developed method demonstrated robustness in diagnosing water deficit stress in complex tree structures.

Conclusions:

• Deep learning regression, combined with frame averaging, offers a precise and robust method for diagnosing water deficit stress in large, field-grown peach trees.
• This image-based approach overcomes the challenges posed by complex tree morphology.
• The technique provides a reliable tool for real-time water status monitoring in orchards.