A Cloud-Based Environment for Generating Yield Estimation Maps From Apple Orchards Using UAV Imagery and a Deep...

Area of Science:

• Agricultural Science
• Computer Vision
• Remote Sensing

Background:

• Accurate crop yield estimation is crucial for agricultural logistics and orchard management.
• Traditional visual yield estimation methods are inaccurate and time-consuming.
• Unmanned Aerial Vehicle (UAV) technology offers potential for efficient agricultural monitoring.

Purpose of the Study:

• To develop a rapid sensing and yield estimation scheme for apple orchards using aerial imagery and deep learning.
• To compare the accuracy of the proposed automated method with manual counts.
• To generate yield maps for optimized orchard management.

Main Methods:

• Training a Region-Convolutional Neural Network (R-CNN) to detect and count apple fruits from UAV-based orthomosaic images.
• Utilizing linear regression to estimate total fruit count per tree, accounting for partially visible fruits.
• Generating geospatial data (shapefiles) and yield maps using Python scripting in Google Colab.

Main Results:

• The R-CNN model achieved a high correlation (R=0.86) with in-situ apple counts.
• Linear regression for total yield estimation yielded a correlation of R=0.80.
• The study successfully generated tree-level and tree-row yield maps.

Conclusions:

• The proposed deep learning approach provides an accurate and efficient method for apple yield estimation.
• Automated yield mapping using UAV imagery can significantly improve orchard management strategies.
• This technology has the potential to enhance crop producer outputs through better planning and resource allocation.