Artificial intelligence-Driven detection and decision support system for precision management of maize downy mildew

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Maize downy mildew (MDM) causes significant crop losses, necessitating early detection.
• Traditional disease detection methods can be time-consuming and less accurate.
• AI offers potential for rapid and precise plant disease identification.

Purpose of the Study:

• To evaluate machine learning (ML) and deep learning (DL) algorithms for maize downy mildew detection.
• To develop an interpretable and accurate AI model for classifying healthy and infected maize leaves.
• To create a decision support system (DSS) for farm-level advisory and disease management.

Main Methods:

• Evaluated thirteen ML/DL algorithms on a field dataset of maize leaf images.
• Utilized VGG16 for its superior performance in classification tasks.
• Employed t-SNE for feature visualization and Grad-CAM for model interpretability.
• Developed a web-based application integrating AI classification with advisory measures.
• Conducted two-year field trials to assess DSS-guided fungicide applications.

Main Results:

• VGG16 achieved high classification accuracy (97%), precision (0.98), recall (0.95), F1-score (0.97), and AUC-ROC (0.99).
• Feature visualization and Grad-CAM confirmed model accuracy and focus on relevant disease symptoms.
• The web-based DSS guided fungicide applications, significantly reducing disease severity and increasing grain yield and economic returns.
• DSS-guided treatments improved yield by 195-289% and economic returns (B:C ratio 3.36-3.57).

Conclusions:

• AI models, particularly VGG16, provide accurate and interpretable solutions for maize downy mildew detection.
• Integrated AI decision support systems enhance precision agriculture, leading to improved crop yields and profitability.
• This approach contributes to sustainable agricultural practices through effective disease management.