Direct UAV-Based Detection of Botrytis cinerea in Vineyards Using Chlorophyll-Absorption Indices and YOLO Deep...

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Agriculture 5.0 necessitates intelligent systems for autonomous crop health monitoring.
• Early and accurate detection of grapevine diseases like Botrytis cinerea is crucial for yield preservation.

Purpose of the Study:

• To design and evaluate an AI-based Unmanned Aerial Vehicle (UAV) system for detecting Botrytis cinerea in vineyards.
• To assess the efficacy of multispectral imagery and deep learning for automated disease identification.

Main Methods:

• Integration of calibrated multispectral data, vegetation indices (specifically CARI), and a YOLOv8 object detection model.
• Field evaluation of the AI-UAV system under real vineyard conditions.
• Comparison of detection performance using multispectral versus RGB imagery.

Main Results:

• The CARI-trained model achieved high detection performance: 92.6% precision, 89.6% recall, 91.1% F1-score, and 93.9% mAP@50.
• Multispectral imagery significantly outperformed RGB imagery (F1-score 68.1%, mAP@50 68.5%).
• The system demonstrated near real-time operation with an average inference time below 50 ms per image.

Conclusions:

• Physiologically informed spectral feature selection (CARI) substantially enhances early Botrytis cinerea detection.
• The proposed UAV-AI framework is suitable for precision viticulture applications in Agriculture 5.0.
• The system enables automated, geolocated disease detection for improved vineyard management.