AI-IoT driven system for agricultural pest outbreak risk prediction

Area of Science:

• Agricultural Science
• Artificial Intelligence
• Computer Vision

Background:

• Invasive pests like the Fall Armyworm (FAW) significantly threaten maize production and global food security.
• Effective pest management necessitates early detection of pest life stages and precise risk prediction for outbreaks.
• Current methods may lack the efficiency and accuracy required for real-time agricultural applications.

Purpose of the Study:

• To develop a hybrid model integrating Explainable Artificial Intelligence (XAI), a lightweight Convolutional Neural Network (CNN), and Fuzzy Logic (FL).
• To achieve accurate detection of FAW and reliable weather-based prediction of pest outbreak risks.
• To create a power-efficient and scalable solution for precision agriculture.

Main Methods:

• Utilized Tiny-MobileNet-SE, a lightweight CNN, for image classification of FAW.
• Employed Grad-CAM for model interpretability, enhancing understanding of detection processes.
• Integrated Fuzzy Logic (FL) for risk prediction using environmental parameters.

Main Results:

• Tiny-MobileNet-SE achieved high performance metrics: 98.6% accuracy, 98.5% F1-score, and 98.6% recall.
• The model demonstrated efficiency with a compact size (0.72 MB) and low latency (80 ms) on Raspberry Pi 5.
• Outperformed other state-of-the-art lightweight models in detection tasks.

Conclusions:

• The proposed hybrid AI system offers a power-efficient, scalable, and user-friendly solution for precision agriculture.
• Provides actionable insights for effective FAW management and supports sustainable crop protection.
• Demonstrates the potential of integrating XAI, CNNs, and FL for advanced agricultural pest monitoring.