A Low-Cost and Unsupervised Image Recognition Methodology for Yield Estimation in a Vineyard

Area of Science:

• Agricultural Science
• Computer Vision
• Remote Sensing

Background:

• Traditional grape yield estimation relies on time-consuming manual sampling, often yielding insufficient data.
• Non-invasive machine vision methods are being explored for rapid and representative yield assessment.

Purpose of the Study:

• To develop an automated system for estimating grape yield using UAV-based RGB imagery.
• To assess grape yield in terms of cluster number and size, considering vigor variability within vineyards.

Main Methods:

• Utilized a low-cost Unmanned Aerial Vehicle (UAV) platform with high-resolution RGB cameras.
• Employed an unsupervised recognition algorithm for cluster detection and size analysis from aerial images.
• Conducted flight campaigns under varying light and canopy cover conditions across two crop seasons.

Main Results:

• Achieved over 85% performance in cluster detection under partially leaf-removed and fully ripe conditions.
• Enabled grapevine yield estimation with more than 84% accuracy several weeks prior to harvest.
• Presented results on cluster number detection and weight estimation across different vineyard vigor zones.

Conclusions:

• Innovative technologies like UAVs, high-resolution cameras, and visual computing offer a novel methodology for grape yield assessment.
• This automated approach significantly saves time and provides accurate yield estimates compared to manual methods.
• The developed system enhances vineyard management and supports achieving desired grape quality.