Determination of Wheat Heading Stage Using Convolutional Neural Networks on Multispectral UAV Imaging Data

Area of Science:

• Agricultural remote sensing
• Plant pathology
• Machine learning in agriculture

Background:

• Accurate monitoring of wheat growth stages, particularly heading and flowering, is crucial for effective fusarium head blight (FHB) management.
• Current manual inspection methods for determining FHB prevention timing are subjective and inefficient, especially in complex terrains.

Purpose of the Study:

• To develop an efficient and accurate method for monitoring wheat growth status during critical stages using unmanned aerial vehicles (UAVs).
• To establish an optimal timing strategy for fusarium head blight (FHB) prevention and control operations.

Main Methods:

• Utilized UAVs equipped with multispectral cameras to capture wheat canopy data during heading and flowering stages.
• Developed and applied a 1D convolutional neural network plus decision tree (1D CNN+DT) model for feature extraction and heading rate regression.
• Integrated regression model outputs into a discrimination model to determine FHB prevention operation timing.

Main Results:

• The 1D CNN+DT model achieved a high coefficient of determination (R² = 0.95) and low root mean square error (RMSE = 0.24) in predicting wheat heading rates from multispectral data.
• The developed discrimination model achieved 97.50% precision in identifying the optimal time for FHB prevention and plant protection operations.
• The proposed UAV-based method significantly outperformed traditional methods and other machine learning models (NN, SVR, DT) in accuracy and efficiency.

Conclusions:

• UAV-based multispectral imaging combined with a 1D CNN+DT model provides a robust and precise method for monitoring wheat growth and optimizing FHB prevention timing.
• This technology offers a significant advancement over manual inspection, enabling more effective and timely crop protection strategies in agriculture.
• The findings support the application of advanced remote sensing and machine learning techniques for precision agriculture and disease management.