In-Season Cotton Yield Prediction with Scale-Aware Convolutional Neural Network Models and Unmanned Aerial Vehicle...

Area of Science:

• Agricultural Science
• Computer Science
• Remote Sensing

Background:

• Efficient water management is vital for sustainable agriculture.
• Traditional cotton yield prediction methods lack accuracy in capturing crop complexities.
• Advanced techniques are needed for informed decision-making in crop management.

Purpose of the Study:

• To develop a novel approach for cotton yield prediction using Unmanned Aerial Vehicles (UAVs) and scale-aware Convolutional Neural Networks (CNNs).
• To leverage spatiotemporal dynamics from high-resolution UAV imagery for enhanced prediction accuracy.
• To improve cotton cultivation practices and overall productivity through precision agriculture.

Main Methods:

• Utilized high-resolution UAV imagery to capture crop health and growth dynamics.
• Employed scale-aware Convolutional Neural Networks (CNNs) to extract spatial and temporal features.
• Conducted experiments with four irrigation treatments on cotton yield at the USDA-ARS CSRL in Lubbock, Texas.

Main Results:

• The proposed CNN regression models outperformed conventional models (AlexNet, CNN-3D, CNN-LSTM, ResNet).
• Achieved state-of-the-art performance with R2 exceeding 0.9 across different image scales.
• Demonstrated low error rates: MAE of 3.08 lbs/row and 7.76% MAPE at the row level; MAE of 0.05 lbs and 10% MAPE at the grid level.

Conclusions:

• Integrating UAV imagery and CNN regression models is a potent strategy for advancing precision agriculture.
• The proposed model effectively captures the interplay of spatial and temporal factors affecting cotton yield.
• This approach offers growers a powerful tool to optimize cultivation and enhance cotton productivity.