Integrating deep learning and field validation into a decision support system for Northern Corn Leaf Blight management in maize

Abstract Northern Corn Leaf Blight (NCLB; also, Turcicum Leaf Blight, TLB), caused by Exserohilum turcicum (teleomorph: Setosphaeria turcica ), is one of the most destructive foliar diseases of maize worldwide, often causing severe yield losses under favorable conditions. We developed a maize-specific, web-based Decision Support System (DSS) for real-time NCLB detection and management ( https://maize-nclb.streamlit.app/ ), integrating advanced deep-learning for automated diagnosis and fungicide advisory. Among thirteen Machine-learning and deep-learning models evaluated for classification, the Visual Geometry Group 16-layer convolutional neural network (VGG16) outperformed all others, achieving 94.0% accuracy, with balanced precision, recall, and F1-score of 0.94, and an AUC-ROC of 0.93. Confusion matrix analysis revealed minimal misclassification, with only 12 errors out of 357 samples, confirming the model’s high reliability in distinguishing healthy and infected plants, while Grad-CAM visualizations consistently highlighted biologically meaningful lesion regions, supporting the model’s interpretability and alignment with plant pathological symptoms. Field validation of DSS-guided fungicide recommendations (Azoxystrobin 18.2% + Difenoconazole 11.4% w/w SC) demonstrated significant benefits, reducing disease incidence to 6.8% compared with 67.4% in controls, achieving 90% disease reduction, and enhancing grain yield by 35.4% (8.55 t/ha), with a favorable cost–benefit ratio of 1:2.49. Seasonal disease progression analysis further confirmed DSS effectiveness, with cumulative disease burden reduced by approximately 85% compared with untreated control. These results highlight the potential of integrating deep-learning with field—validated management strategies into a practical DSS, demonstrating its potential for precision disease management in maize.