Tree-level almond yield estimation from high resolution aerial imagery with convolutional neural network

Area of Science:

• Agricultural Science
• Computer Science
• Remote Sensing

Background:

• Precision agriculture requires accurate yield variability estimation for high-value tree crops.
• Advancements in sensor technology and machine learning enable high-resolution orchard monitoring and tree-level yield prediction.

Purpose of the Study:

• To evaluate deep learning methods for predicting tree-level almond yield using multi-spectral imagery.
• To assess the efficacy of a Convolutional Neural Network (CNN) with a spatial attention module for fresh weight estimation.

Main Methods:

• A CNN model was developed using multi-spectral aerial imagery (30cm resolution, 4 bands) from an 'Independence' almond orchard in California.
• Individual tree yield data from ~2,000 trees were collected and used for model training and validation.
• The model directly processed reflectance imagery for tree-level fresh weight estimation.

Main Results:

• The CNN model achieved high accuracy, with an R-squared of 0.96 (±0.002) and Normalized Root Mean Square Error (NRMSE) of 6.6% (±0.2%).
• The model effectively captured spatial yield variations within the orchard at row, transect, and individual tree levels.
• The red edge spectral band was identified as the most influential for yield estimation.

Conclusions:

• Deep learning significantly outperforms traditional methods for accurate and robust tree-level yield estimation.
• This data-driven approach supports site-specific resource management for enhanced agricultural sustainability.
• The study highlights the potential of advanced machine learning and remote sensing in modern agriculture.