Recovering Wind-Induced Plant Motion in Dense Field Environments via Deep Learning and Multiple Object Tracking

Area of Science:

• Agricultural Science
• Plant Biology
• Computer Vision

Background:

• Understanding plant responses to environmental conditions is crucial for crop improvement.
• Wind-induced motion in crops is vital but understudied due to field complexities.
• Analyzing plant movement in dense stands is challenging with traditional methods.

Purpose of the Study:

• To develop a robust method for characterizing motion in field-grown wheat plants (Triticum aestivum).
• To enable automated detection and tracking of wheat ears for movement analysis.
• To provide data for assessing plant structure, function, and informing agricultural models.

Main Methods:

• Utilized time-ordered RGB image sequences of wheat plants.
• Trained a convolutional neural network on annotated ear tip images with data augmentation.
• Applied a probabilistic tracking algorithm to follow ear tips across video frames.

Main Results:

• Successfully detected wheat ears in the field without camera calibration or fixed positions.
• Tracked ear tip movement to approximate plant motion dynamics.
• Enabled characterization of movement traits like periodicity and static plant properties.

Conclusions:

• The developed method offers a robust way to study wind-induced plant motion in field conditions.
• Automated data extraction can inform crop lodging models and breeding programs.
• Linking movement properties to light distribution can enhance understanding of plant function.