Agentic AI for smart and sustainable precision agriculture

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Modern agriculture faces challenges in achieving sustainable and efficient practices.
• Precision Agriculture (PA) and Federated Learning (FL) offer potential solutions for optimizing resource use and decision-making.

Purpose of the Study:

• To propose and evaluate an Agentic Artificial Intelligence (AAI) framework for precision agriculture.
• To integrate distributed sensing, intelligent agents, and FL for real-time farm-level monitoring and decision support.

Main Methods:

• Developed an AAI framework integrating PA and FL with local models (DenseNet121, MobileNetV2, EfficientDet-D0, YOLOv8).
• Evaluated the framework on tomato disease classification and weed detection datasets.
• Outlined a practical deployment architecture for inter-device communication and localized intelligence.

Main Results:

• The federated global model achieved 96.4% accuracy, surpassing individual client models.
• EfficientDet-D0 achieved high performance in weed detection (mAP@0.5: 0.978, F1-score: 0.961).
• DenseNet121 and MobileNetV2 showed strong performance in classification tasks (95.0% and 93.9% accuracy).

Conclusions:

• The integration of AAI and FL is feasible and effective for intelligent precision agriculture.
• The study provides a roadmap for developing sustainable intelligent farming systems.
• SWOT analysis identified strengths and deployment challenges, guiding future research.