An Ensemble Learning Model for Detecting Soybean Seedling Emergence in UAV Imagery

Area of Science:

• Agricultural Science
• Computer Science
• Remote Sensing

Background:

• Accurate soybean seedling emergence assessment is crucial for effective field management.
• Current methods using multiple models are slow and hinder timely decision-making.

Purpose of the Study:

• To develop an integrated deep learning model for evaluating multiple soybean seedling emergence indicators.
• To improve the efficiency and accuracy of soybean emergence detection and uniformity assessment.

Main Methods:

• Utilized unmanned aerial vehicle (UAV) to capture RGB images at different soybean growth stages (VE, VC, V1).
• Implemented a seedling emergence detection module and an automatic seedling cutting module for dataset construction.
• Employed an improved AlexNet as the backbone for the growth stage discrimination module.

Main Results:

• The seedling emergence detection module achieved high accuracy (99.92%) and strong correlation (R²=0.9784).
• The improved AlexNet demonstrated reduced training time and high accuracy (99.07%) with low loss (0.0355).
• Field validation showed minimal error (0.0060 to 0.0775) between predicted and actual emergence proportions.

Conclusions:

• The developed ensemble learning model effectively detects and evaluates soybean seedling emergence and uniformity.
• This approach provides a theoretical basis for informed soybean field management and precision operations.
• The model shows potential for evaluating emergence information in other crop types.