Assessment of Soybean Lodging Using UAV Imagery and Machine Learning

Area of Science:

• Agricultural Science
• Plant Breeding
• Remote Sensing

Background:

• Plant lodging is a critical soybean phenotype for breeding, traditionally assessed visually with limitations.
• Visual assessment is time-consuming and prone to human error, necessitating advanced methods.

Purpose of the Study:

• To explore the potential of unmanned aerial vehicle (UAV)-based imagery and machine learning for soybean lodging assessment.
• To develop an automated system for classifying soybean lodging severity in breeding programs.

Main Methods:

• Collected UAV RGB imagery of 1266 soybean plots at the reproductive stage.
• Segmented plots and extracted 12 image features for lodging assessment.
• Evaluated four machine learning models (XGBoost, RF, KNN, ANN) with SMOTE-ENN preprocessing for imbalanced data.

Main Results:

• The Synthetic Minority Oversampling-Edited Nearest Neighbor (SMOTE-ENN) preprocessing method improved classification accuracy across all models.
• The artificial neural network (ANN) classifier achieved 96% overall accuracy when using the SMOTE-ENN processed dataset.
• The study demonstrated the effectiveness of UAV imagery and machine learning in differentiating soybean lodging phenotypes.

Conclusions:

• UAV-based imagery and machine learning offer a robust and accurate method for soybean lodging assessment.
• The developed classification model can be integrated into breeding programs for efficient phenotype evaluation.
• SMOTE-ENN is a suitable preprocessing technique for imbalanced datasets in this classification task.