Framework for smartphone-based grape detection and vineyard management using UAV-trained AI.

Area of Science:

• Agricultural Technology
• Computer Vision
• Machine Learning

Background:

• Viticulture requires efficient grape bunch identification for yield and quality assessment.
• Traditional methods are labor-intensive, while advanced drone-based systems can be inaccessible to farmers.
• Smartphones offer a widely accessible platform for agricultural data collection.

Purpose of the Study:

• To develop an accessible and accurate AI-based system for automated grape bunch detection in vineyards.
• To integrate Unmanned Aerial Vehicle (UAV) data with smartphone imaging for robust model training and deployment.
• To create a practical, affordable, and scalable solution for farmers to monitor grape yield.

Main Methods:

• An AI pipeline combining object detection (YOLO) and pixel-level segmentation (X-Decoder) was developed.
• Unmanned Aerial Vehicle (UAV) videos were used for initial model training and segmentation.
• The trained model was applied to images captured by common smartphones (Xiaomi Poco X3 Pro, iPhone X).
• A web application was created to facilitate system integration with mobile technology.

Main Results:

• The AI system achieved high detection accuracy with a precision of 0.92, recall of 0.735, and F1 score of 0.82.
• The model demonstrated robustness, with AI-detected grape bunches correlating strongly with ground truth (R² = 0.84).
• The integrated approach surpassed traditional and purely UAV-based methods in efficiency and adaptability.

Conclusions:

• Combining UAV data for training and smartphone imaging for application offers a practical and scalable solution for viticulture monitoring.
• The developed AI framework makes advanced grape bunch detection accessible to farmers through readily available technology.
• This approach significantly reduces the time and effort required for vineyard yield assessment, enhancing agricultural practices.