Digital decision support integrated with diagnostics and precision fungicide application for Southern Corn Leaf Blight in maize

Abstract

Abstract Southern Corn Leaf Blight (SCLB, also called Maize Leaf Blight, MLB), caused by Bipolaris maydis (teleomorph: Cochliobolus heterostrophus ), severely limits maize yield under favourable conditions. Rapid detection and precise interventions are essential for sustainable production. We present an AI-driven framework integrating deep learning diagnostics, precision fungicide application, and a digital decision support system (DSS) for field-level SCLB management. Thirteen machine learning (ML) and deep learning (DL) algorithms were evaluated, with VGG16 achieving the highest performance (accuracy 97.0%, precision 0.98, recall 0.96, F1-score ≥ 0.97, AUC-ROC = 1.00). Feature extraction analysis highlighted VGG16’s ability to capture hierarchical disease-specific patterns (score = 0.95), and error- and variance-based assessment confirmed minimal prediction errors (MAE = 0.06, RMSE = 0.16, Explained Variance = 0.90, MBD = − 0.02). Confusion matrix analysis revealed only a small number of misclassifications (4 false negatives and 9 false positives), demonstrating excellent generalization. Grad-CAM heatmaps, t-SNE visualization, and learning curves confirmed lesion-focused predictions and feature separability. Two-year field trials (2023 and 2024) validated precision fungicide application (Azoxystrobin 18.2% + Difenoconazole 11.4% SC), reducing disease severity to ≈ 10% PDI (86.2% reduction) and increasing grain yield to 83.7 q/ha (C: B ratio 1:2.41). The Streamlit-based DSS provides actionable, real-time advisories, offering a scalable AI platform for automated disease detection and precision agriculture in maize. The proposed framework can be extended to other foliar diseases and integrated with IoT-based sensing for region-wide advisory systems.