A Deep Learning Approach for Precision Viticulture, Assessing Grape Maturity via YOLOv7

Area of Science:

• Agricultural Engineering
• Computer Vision
• Viticulture

Background:

• Robots are increasingly used in viticulture to enhance productivity and precision.
• Labor shortages and high costs drive the need for automated vineyard operations.

Purpose of the Study:

• To develop an algorithm for grape maturity estimation in vineyard management.
• To improve automated decision-making for grape harvesting and quality control.

Main Methods:

• Development of an object detection algorithm using You Only Look Once (YOLO) v7.
• Training the algorithm with images of Assyrtiko grapes over six weeks in Drama, Greece.
• Comparing YOLO v7 performance against alternative object detection architectures.

Main Results:

• The YOLO v7 algorithm successfully detected five distinct grape maturity stages.
• The proposed algorithm demonstrated superior precision and accuracy compared to other methods.
• High-quality images were utilized for robust algorithm validation.

Conclusions:

• The developed algorithm enables accurate, automated grape maturity estimation.
• This research supports the advancement of autonomous robots for comprehensive grapevine management.
• The findings have significant implications for optimizing viticulture practices.