Convolutional Neural Network Models Help Effectively Estimate Legume Coverage in Grass-Legume Mixed Swards

Area of Science:

• Agricultural Science
• Computer Science
• Remote Sensing

Background:

• Accurate estimation of legume proportion in grass-legume swards is crucial for forage breeding and cultivation.
• Traditional methods for assessing legume coverage can be time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate convolutional neural network (CNN) models for objective and efficient estimation of legume coverage in mixed swards.
• To compare the accuracy of CNN models trained on different datasets.

Main Methods:

• Fine-tuning GoogLeNet architecture for image analysis.
• Training CNN models on unmanned aerial vehicle (UAV)-based images to estimate coverage of timothy (TY), white clover (WC), and background (Bg).
• Evaluating model accuracy using mean bias error and mean average error.

Main Results:

• CNN models demonstrated high accuracy in estimating legume coverage, with correlation coefficients (r) of 0.92-0.96 for aerial image measurements.
• Models trained on diverse datasets (multiple plots) outperformed those trained on single plots.
• White clover coverage was estimated more precisely than timothy or background.

Conclusions:

• CNN models provide a reliable and effective method for estimating legume coverage in mixed swards.
• The developed models can significantly aid in forage research and breeding programs.
• UAV-based imagery combined with CNNs offers a powerful tool for agricultural monitoring.